Update TODO.md

Tweaks/cleanups to conversation management in tui
- Pass around message/conversation values instead of pointers where it makes more sense, and store values instead of pointers in the globally (within the TUI) shared `App` (pointers provide no utility here). - Split conversation persistence into separate conversation/message saving stages
2024-10-27 18:03:15 +00:00 · 2024-10-25 16:57:15 +00:00 · 2024-10-22 17:53:13 +00:00 · 2024-10-22 17:53:13 +00:00 · 2024-10-22 17:53:13 +00:00 · 2024-10-20 22:39:35 +00:00
63 changed files with 6123 additions and 2742 deletions
--- a/README.md
+++ b/README.md
@ -1,59 +1,174 @@
-# lmcli
+# lmcli - Large ____ Model CLI

-`lmcli` is a (Large) Language Model CLI.
+`lmcli` is a versatile command-line interface for interacting with LLMs and LMMs.

-Current features:
- Perform one-shot prompts with `lmcli prompt <message>`
- Manage persistent conversations with the `new`, `reply`, `view`, `rm`,
-  `edit`, `retry`, `continue` sub-commands.
- Syntax highlighted output
- Tool calling, see the [Tools](#tools) section.
+## Features

-Maybe features:
- Chat-like interface (`lmcli chat`) for rapid back-and-forth conversations
- Support for additional models/APIs besides just OpenAI
+- Multiple model backends (Ollama, OpenAI, Anthropic, Google)
+- Customizable agents with tool calling
+- Persistent conversation management
+- Message branching (edit and re-prompt to your heart's desire)
+- Interactive terminal interface for seamless chat experiences
+- Utilizes `$EDITOR`, write and edit prompts from the comfort of your own editor :)
+- `vi`-like bindings
+- Syntax highlighting!
+
+## Screenshots
+
+[TODO: Add screenshots of the TUI in action, showing different views and features]
+
+## Installation
+
+To install `lmcli`, make sure you have Go installed on your system, then run:
+
+```sh
+go install git.mlow.ca/mlow/lmcli@latest
+```
+
+## Configuration
+
+`lmcli` uses a YAML configuration file located at `~/.config/lmcli/config.yaml`. Here's a sample configuration:
+
+```yaml
+defaults:
+  model: claude-3-5-sonnet-20240620
+  maxTokens: 3072
+  temperature: 0.2
+conversations:
+  titleGenerationModel: claude-3-haiku-20240307
+chroma:
+  style: onedark
+  formatter: terminal16m
+agents:
+  - name: coder
+    tools:
+      - dir_tree
+      - read_file
+      #- write_file
+    systemPrompt: |
+      You are an experienced software engineer...
+  # ...
+providers:
+  - kind: ollama
+    models:
+      - phi3:instruct
+      - llama3:8b
+  - kind: anthropic
+    apiKey: your-api-key-here
+    models:
+      - claude-3-5-sonnet-20240620
+      - claude-3-opus-20240229
+      - claude-3-haiku-20240307
+  - kind: openai
+    apiKey: your-api-key-here
+    models:
+      - gpt-4o
+      - gpt-4-turbo
+  - name: openrouter
+    kind: openai
+    apiKey: your-api-key-here
+    baseUrl: https://openrouter.ai/api/
+    models:
+      - qwen/qwen-2-72b-instruct
+  # ...
+```
+
+Customize this file to add your own providers, agents, and models.
+
+### Syntax highlighting
+
+Syntax highlighting is performed by [Chroma](https://github.com/alecthomas/chroma).
+
+Refer to [`Chroma/styles`](https://github.com/alecthomas/chroma/tree/master/styles) for available styles (TODO: add support for custom Chroma styles).
+
+Available formatters:
+- `terminal` - 8 colors
+- `terminal16` - 16 colors
+- `terminal256` - 256 colors
+- `terminal16m` - true color (default)
+
+## Agents
+
+Agents in `lmcli` combine a system prompt with a set of available tools. Agents are defined in `config.yaml` and are called upon with the `-a`/`--agent` flag.
+
+Agent functionality is expected to be expanded on, bringing them to close parity with something like OpenAI's "Assistants" feature.

 ## Tools
-Tools must be explicitly enabled by adding the tool's name to the
-`openai.enabledTools` array in `config.yaml`.

-Note: all filesystem related tools operate relative to the current directory
-only. They do not accept absolute paths, and efforts are made to ensure they
-cannot escape above the working directory). **Close attention must be paid to
-where you are running `lmcli`, as the model could at any time decide to use one
-of these tools to discover and read potentially sensitive information from your
-filesystem.**
+Tools are used by agents to acquire information from and interact with external systems. The following built-in tools are available:

-It's best to only have tools enabled in `config.yaml` when you intend to be
-using them, since their descriptions (see `pkg/cli/functions.go`) count towards
-context usage.
+- `dir_tree`: Display a directory structure
+- `read_file`: Read the contents of a file
+- `write_file`: Write content to a file
+- `file_insert_lines`: Insert lines at a specific position in a file
+- `file_replace_lines`: Replace a range of lines in a file

-Available tools:
+Obviously, some of these tools carry significant risk. Use wisely :)

- `read_dir` - Read the contents of a directory.
- `read_file` - Read the contents of a file.
- `write_file` - Write contents to a file.
- `file_insert_lines` - Insert lines at a position within a file. Tricky for
-  the model to use, but can potentially save tokens.
- `file_replace_lines` - Remove or replace a range of lines within a file. Even
-  trickier for the model to use.
-
-## Install
-
-```shell
-$ go install git.mlow.ca/mlow/lmcli@latest
-```
+More tool features are planned, including the ability to define arbitrary tools which call out to external scripts, tools to spawn sub-agents, perform web searches, etc.

 ## Usage

-Invoke `lmcli` at least once:
-
-```shell
+```console
 $ lmcli help
+lmcli - Large Language Model CLI
+
+Usage:
+  lmcli <command> [flags]
+  lmcli [command]
+
+Available Commands:
+  chat        Open the chat interface
+  clone       Clone conversations
+  completion  Generate the autocompletion script for the specified shell
+  continue    Continue a conversation from the last message
+  edit        Edit the last user reply in a conversation
+  help        Help about any command
+  list        List conversations
+  new         Start a new conversation
+  prompt      Do a one-shot prompt
+  rename      Rename a conversation
+  reply       Reply to a conversation
+  retry       Retry the last user reply in a conversation
+  rm          Remove conversations
+  view        View messages in a conversation
+
+Flags:
+  -h, --help   help for lmcli
+
+Use "lmcli [command] --help" for more information about a command.
 ```

-Edit `~/.config/lmcli/config.yaml` and set `openai.apiKey` to your API key.
+### Examples

-Refer back to the output of `lmcli help` for usage.
+Start a new chat with the `coder` agent:

-Enjoy!
+```console
+$ lmcli chat --agent coder
+```
+
+Start a new conversation, imperative style (no tui):
+
+```console
+$ lmcli new "Help me plan meals for the next week"
+```
+
+Send a one-shot prompt (no persistence):
+
+```console
+$ lmcli prompt "What is the answer to life, the universe, and everything?"
+```
+
+## Contributing
+
+Contributions to `lmcli` are welcome! Feel free to open issues or submit pull requests on the project repository.
+
+For a full list of planned features and improvements, check out the [TODO.md](TODO.md) file.
+
+## License
+
+To be determined
+
+## Acknowledgements
+
+`lmcli` is a small hobby project. Special thanks to the Go community and the creators of the libraries used in this project.
--- a/TODO.md
+++ b/TODO.md
@ -0,0 +1,37 @@
+# TODO
+
+- [x] Strip anthropic XML function call scheme from content, to reconstruct
+  when calling anthropic?
+- [x] `dir_tree` tool
+- [x] Implement native Anthropic API tool calling
+- [x] Agents - a name given to a system prompt + set of available tools +
+  potentially other relevent data (e.g. external service credentials, files for
+  RAG, etc), which the user explicitly selects (e.g. `lmcli chat --agent
+  code-helper`, `lmcli chat -a financier`).
+    - [ ] Specialized agents which have integrations beyond basic tool calling,
+      e.g. a coding agent which bakes in efficient code context management
+      (only the current state of relevant files get shown to the model in the
+      system prompt, rather than having them in the conversation messages)
+    - [ ] Agents may have some form of long term memory management (key-value?
+      natural lang?).
+- [ ] Sandboxed python, js interpreters (implemented with containers)
+- [ ] Support for arbitrary external script tools
+- [ ] Search - RAG driven search of existing conversation "hey, remind me of
+  the conversation we had six months ago about X")
+- [ ] Conversation categorization - model driven category creation and
+  conversation classification
+- [ ] Image input
+- [ ] Image output (sixel support?)
+- [ ] Conversation exports to html/pdf/json
+- [ ] Store message generation model
+- [ ] Hidden CoT
+- [ ] Token accounting
+
+## UI
+- [x] Prettify/normalize tool_call and tool_result outputs so they can be
+  shown/optionally hidden in `lmcli view` and `lmcli chat`
+- [x] Conversation deletion in conversations view
+- [ ] User confirmation before calling (some?) tools
+- [ ] Message deletion, Ctrl+D to delete a message and attach its children to
+  its parent, Ctrl+Shift+D to delete a message and its descendents
+- [ ] Show available key bindings and their action in any given view
--- a/go.mod
+++ b/go.mod
@ -3,42 +3,41 @@ module git.mlow.ca/mlow/lmcli
 go 1.21

 require (
-	github.com/alecthomas/chroma/v2 v2.11.1
-	github.com/charmbracelet/bubbles v0.18.0
-	github.com/charmbracelet/bubbletea v0.25.0
-	github.com/charmbracelet/lipgloss v0.10.0
-	github.com/go-yaml/yaml v2.1.0+incompatible
-	github.com/sashabaranov/go-openai v1.17.7
-	github.com/spf13/cobra v1.8.0
+	github.com/alecthomas/chroma/v2 v2.14.0
+	github.com/charmbracelet/bubbles v0.20.0
+	github.com/charmbracelet/bubbletea v1.1.1
+	github.com/charmbracelet/lipgloss v0.13.0
+	github.com/muesli/reflow v0.3.0
+	github.com/spf13/cobra v1.8.1
 	github.com/sqids/sqids-go v0.4.1
-	gorm.io/driver/sqlite v1.5.4
-	gorm.io/gorm v1.25.5
+	gopkg.in/yaml.v3 v3.0.1
+	gorm.io/driver/sqlite v1.5.6
+	gorm.io/gorm v1.25.12
 )

 require (
 	github.com/atotto/clipboard v0.1.4 // indirect
 	github.com/aymanbagabas/go-osc52/v2 v2.0.1 // indirect
-	github.com/containerd/console v1.0.4-0.20230313162750-1ae8d489ac81 // indirect
-	github.com/dlclark/regexp2 v1.10.0 // indirect
+	github.com/charmbracelet/x/ansi v0.3.1 // indirect
+	github.com/charmbracelet/x/term v0.2.0 // indirect
+	github.com/dlclark/regexp2 v1.11.4 // indirect
+	github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f // indirect
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
 	github.com/jinzhu/inflection v1.0.0 // indirect
 	github.com/jinzhu/now v1.1.5 // indirect
 	github.com/kr/pretty v0.3.1 // indirect
 	github.com/lucasb-eyer/go-colorful v1.2.0 // indirect
-	github.com/mattn/go-isatty v0.0.18 // indirect
+	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/mattn/go-localereader v0.0.1 // indirect
-	github.com/mattn/go-runewidth v0.0.15 // indirect
-	github.com/mattn/go-sqlite3 v1.14.18 // indirect
-	github.com/muesli/ansi v0.0.0-20211018074035-2e021307bc4b // indirect
+	github.com/mattn/go-runewidth v0.0.16 // indirect
+	github.com/mattn/go-sqlite3 v1.14.23 // indirect
+	github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 // indirect
 	github.com/muesli/cancelreader v0.2.2 // indirect
-	github.com/muesli/reflow v0.3.0 // indirect
 	github.com/muesli/termenv v0.15.2 // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/spf13/pflag v1.0.5 // indirect
-	golang.org/x/sync v0.1.0 // indirect
-	golang.org/x/sys v0.14.0 // indirect
-	golang.org/x/term v0.6.0 // indirect
-	golang.org/x/text v0.3.8 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.25.0 // indirect
+	golang.org/x/text v0.18.0 // indirect
 	gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15 // indirect
-	gopkg.in/yaml.v2 v2.2.2 // indirect
 )
--- a/go.sum
+++ b/go.sum
@ -1,27 +1,31 @@
-github.com/alecthomas/assert/v2 v2.2.1 h1:XivOgYcduV98QCahG8T5XTezV5bylXe+lBxLG2K2ink=
-github.com/alecthomas/assert/v2 v2.2.1/go.mod h1:pXcQ2Asjp247dahGEmsZ6ru0UVwnkhktn7S0bBDLxvQ=
-github.com/alecthomas/chroma/v2 v2.11.1 h1:m9uUtgcdAwgfFNxuqj7AIG75jD2YmL61BBIJWtdzJPs=
-github.com/alecthomas/chroma/v2 v2.11.1/go.mod h1:4TQu7gdfuPjSh76j78ietmqh9LiurGF0EpseFXdKMBw=
-github.com/alecthomas/repr v0.2.0 h1:HAzS41CIzNW5syS8Mf9UwXhNH1J9aix/BvDRf1Ml2Yk=
-github.com/alecthomas/repr v0.2.0/go.mod h1:Fr0507jx4eOXV7AlPV6AVZLYrLIuIeSOWtW57eE/O/4=
+github.com/MakeNowJust/heredoc v1.0.0 h1:cXCdzVdstXyiTqTvfqk9SDHpKNjxuom+DOlyEeQ4pzQ=
+github.com/MakeNowJust/heredoc v1.0.0/go.mod h1:mG5amYoWBHf8vpLOuehzbGGw0EHxpZZ6lCpQ4fNJ8LE=
+github.com/alecthomas/assert/v2 v2.7.0 h1:QtqSACNS3tF7oasA8CU6A6sXZSBDqnm7RfpLl9bZqbE=
+github.com/alecthomas/assert/v2 v2.7.0/go.mod h1:Bze95FyfUr7x34QZrjL+XP+0qgp/zg8yS+TtBj1WA3k=
+github.com/alecthomas/chroma/v2 v2.14.0 h1:R3+wzpnUArGcQz7fCETQBzO5n9IMNi13iIs46aU4V9E=
+github.com/alecthomas/chroma/v2 v2.14.0/go.mod h1:QolEbTfmUHIMVpBqxeDnNBj2uoeI4EbYP4i6n68SG4I=
+github.com/alecthomas/repr v0.4.0 h1:GhI2A8MACjfegCPVq9f1FLvIBS+DrQ2KQBFZP1iFzXc=
+github.com/alecthomas/repr v0.4.0/go.mod h1:Fr0507jx4eOXV7AlPV6AVZLYrLIuIeSOWtW57eE/O/4=
 github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z4=
 github.com/atotto/clipboard v0.1.4/go.mod h1:ZY9tmq7sm5xIbd9bOK4onWV4S6X0u6GY7Vn0Yu86PYI=
 github.com/aymanbagabas/go-osc52/v2 v2.0.1 h1:HwpRHbFMcZLEVr42D4p7XBqjyuxQH5SMiErDT4WkJ2k=
 github.com/aymanbagabas/go-osc52/v2 v2.0.1/go.mod h1:uYgXzlJ7ZpABp8OJ+exZzJJhRNQ2ASbcXHWsFqH8hp8=
-github.com/charmbracelet/bubbles v0.18.0 h1:PYv1A036luoBGroX6VWjQIE9Syf2Wby2oOl/39KLfy0=
-github.com/charmbracelet/bubbles v0.18.0/go.mod h1:08qhZhtIwzgrtBjAcJnij1t1H0ZRjwHyGsy6AL11PSw=
-github.com/charmbracelet/bubbletea v0.25.0 h1:bAfwk7jRz7FKFl9RzlIULPkStffg5k6pNt5dywy4TcM=
-github.com/charmbracelet/bubbletea v0.25.0/go.mod h1:EN3QDR1T5ZdWmdfDzYcqOCAps45+QIJbLOBxmVNWNNg=
-github.com/charmbracelet/lipgloss v0.10.0 h1:KWeXFSexGcfahHX+54URiZGkBFazf70JNMtwg/AFW3s=
-github.com/charmbracelet/lipgloss v0.10.0/go.mod h1:Wig9DSfvANsxqkRsqj6x87irdy123SR4dOXlKa91ciE=
-github.com/containerd/console v1.0.4-0.20230313162750-1ae8d489ac81 h1:q2hJAaP1k2wIvVRd/hEHD7lacgqrCPS+k8g1MndzfWY=
-github.com/containerd/console v1.0.4-0.20230313162750-1ae8d489ac81/go.mod h1:YynlIjWYF8myEu6sdkwKIvGQq+cOckRm6So2avqoYAk=
-github.com/cpuguy83/go-md2man/v2 v2.0.3/go.mod h1:tgQtvFlXSQOSOSIRvRPT7W67SCa46tRHOmNcaadrF8o=
+github.com/charmbracelet/bubbles v0.20.0 h1:jSZu6qD8cRQ6k9OMfR1WlM+ruM8fkPWkHvQWD9LIutE=
+github.com/charmbracelet/bubbles v0.20.0/go.mod h1:39slydyswPy+uVOHZ5x/GjwVAFkCsV8IIVy+4MhzwwU=
+github.com/charmbracelet/bubbletea v1.1.1 h1:KJ2/DnmpfqFtDNVTvYZ6zpPFL9iRCRr0qqKOCvppbPY=
+github.com/charmbracelet/bubbletea v1.1.1/go.mod h1:9Ogk0HrdbHolIKHdjfFpyXJmiCzGwy+FesYkZr7hYU4=
+github.com/charmbracelet/lipgloss v0.13.0 h1:4X3PPeoWEDCMvzDvGmTajSyYPcZM4+y8sCA/SsA3cjw=
+github.com/charmbracelet/lipgloss v0.13.0/go.mod h1:nw4zy0SBX/F/eAO1cWdcvy6qnkDUxr8Lw7dvFrAIbbY=
+github.com/charmbracelet/x/ansi v0.3.1 h1:CRO6lc/6HCx2/D6S/GZ87jDvRvk6GtPyFP+IljkNtqI=
+github.com/charmbracelet/x/ansi v0.3.1/go.mod h1:dk73KoMTT5AX5BsX0KrqhsTqAnhZZoCBjs7dGWp4Ktw=
+github.com/charmbracelet/x/term v0.2.0 h1:cNB9Ot9q8I711MyZ7myUR5HFWL/lc3OpU8jZ4hwm0x0=
+github.com/charmbracelet/x/term v0.2.0/go.mod h1:GVxgxAbjUrmpvIINHIQnJJKpMlHiZ4cktEQCN6GWyF0=
+github.com/cpuguy83/go-md2man/v2 v2.0.4/go.mod h1:tgQtvFlXSQOSOSIRvRPT7W67SCa46tRHOmNcaadrF8o=
 github.com/creack/pty v1.1.9/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ33E=
-github.com/dlclark/regexp2 v1.10.0 h1:+/GIL799phkJqYW+3YbOd8LCcbHzT0Pbo8zl70MHsq0=
-github.com/dlclark/regexp2 v1.10.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
-github.com/go-yaml/yaml v2.1.0+incompatible h1:RYi2hDdss1u4YE7GwixGzWwVo47T8UQwnTLB6vQiq+o=
-github.com/go-yaml/yaml v2.1.0+incompatible/go.mod h1:w2MrLa16VYP0jy6N7M5kHaCkaLENm+P+Tv+MfurjSw0=
+github.com/dlclark/regexp2 v1.11.4 h1:rPYF9/LECdNymJufQKmri9gV604RvvABwgOA8un7yAo=
+github.com/dlclark/regexp2 v1.11.4/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
+github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f h1:Y/CXytFA4m6baUTXGLOoWe4PQhGxaX0KpnayAqC48p4=
+github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f/go.mod h1:vw97MGsxSvLiUE2X8qFplwetxpGLQrlU1Q9AUEIzCaM=
 github.com/hexops/gotextdiff v1.0.3 h1:gitA9+qJrrTCsiCl7+kh75nPqQt1cx4ZkudSTLoUqJM=
 github.com/hexops/gotextdiff v1.0.3/go.mod h1:pSWU5MAI3yDq+fZBTazCSJysOMbxWL1BSow5/V2vxeg=
 github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
@ -36,17 +40,17 @@ github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
 github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/lucasb-eyer/go-colorful v1.2.0 h1:1nnpGOrhyZZuNyfu1QjKiUICQ74+3FNCN69Aj6K7nkY=
 github.com/lucasb-eyer/go-colorful v1.2.0/go.mod h1:R4dSotOR9KMtayYi1e77YzuveK+i7ruzyGqttikkLy0=
-github.com/mattn/go-isatty v0.0.18 h1:DOKFKCQ7FNG2L1rbrmstDN4QVRdS89Nkh85u68Uwp98=
-github.com/mattn/go-isatty v0.0.18/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
+github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
 github.com/mattn/go-localereader v0.0.1 h1:ygSAOl7ZXTx4RdPYinUpg6W99U8jWvWi9Ye2JC/oIi4=
 github.com/mattn/go-localereader v0.0.1/go.mod h1:8fBrzywKY7BI3czFoHkuzRoWE9C+EiG4R1k4Cjx5p88=
 github.com/mattn/go-runewidth v0.0.12/go.mod h1:RAqKPSqVFrSLVXbA8x7dzmKdmGzieGRCM46jaSJTDAk=
-github.com/mattn/go-runewidth v0.0.15 h1:UNAjwbU9l54TA3KzvqLGxwWjHmMgBUVhBiTjelZgg3U=
-github.com/mattn/go-runewidth v0.0.15/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
-github.com/mattn/go-sqlite3 v1.14.18 h1:JL0eqdCOq6DJVNPSvArO/bIV9/P7fbGrV00LZHc+5aI=
-github.com/mattn/go-sqlite3 v1.14.18/go.mod h1:2eHXhiwb8IkHr+BDWZGa96P6+rkvnG63S2DGjv9HUNg=
-github.com/muesli/ansi v0.0.0-20211018074035-2e021307bc4b h1:1XF24mVaiu7u+CFywTdcDo2ie1pzzhwjt6RHqzpMU34=
-github.com/muesli/ansi v0.0.0-20211018074035-2e021307bc4b/go.mod h1:fQuZ0gauxyBcmsdE3ZT4NasjaRdxmbCS0jRHsrWu3Ho=
+github.com/mattn/go-runewidth v0.0.16 h1:E5ScNMtiwvlvB5paMFdw9p4kSQzbXFikJ5SQO6TULQc=
+github.com/mattn/go-runewidth v0.0.16/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
+github.com/mattn/go-sqlite3 v1.14.23 h1:gbShiuAP1W5j9UOksQ06aiiqPMxYecovVGwmTxWtuw0=
+github.com/mattn/go-sqlite3 v1.14.23/go.mod h1:Uh1q+B4BYcTPb+yiD3kU8Ct7aC0hY9fxUwlHK0RXw+Y=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 h1:ZK8zHtRHOkbHy6Mmr5D264iyp3TiX5OmNcI5cIARiQI=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6/go.mod h1:CJlz5H+gyd6CUWT45Oy4q24RdLyn7Md9Vj2/ldJBSIo=
 github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELUXHmA=
 github.com/muesli/cancelreader v0.2.2/go.mod h1:3XuTXfFS2VjM+HTLZY9Ak0l6eUKfijIfMUZ4EgX0QYo=
 github.com/muesli/reflow v0.3.0 h1:IFsN6K9NfGtjeggFP+68I4chLZV2yIKsXJFNZ+eWh6s=
@ -61,31 +65,26 @@ github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUc
 github.com/rogpeppe/go-internal v1.9.0 h1:73kH8U+JUqXU8lRuOHeVHaa/SZPifC7BkcraZVejAe8=
 github.com/rogpeppe/go-internal v1.9.0/go.mod h1:WtVeX8xhTBvf0smdhujwtBcq4Qrzq/fJaraNFVN+nFs=
 github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
-github.com/sashabaranov/go-openai v1.17.7 h1:MPcAwlwbeo7ZmhQczoOgZBHtIBY1TfZqsdx6+/ndloM=
-github.com/sashabaranov/go-openai v1.17.7/go.mod h1:lj5b/K+zjTSFxVLijLSTDZuP7adOgerWeFyZLUhAKRg=
-github.com/spf13/cobra v1.8.0 h1:7aJaZx1B85qltLMc546zn58BxxfZdR/W22ej9CFoEf0=
-github.com/spf13/cobra v1.8.0/go.mod h1:WXLWApfZ71AjXPya3WOlMsY9yMs7YeiHhFVlvLyhcho=
+github.com/spf13/cobra v1.8.1 h1:e5/vxKd/rZsfSJMUX1agtjeTDf+qv1/JdBF8gg5k9ZM=
+github.com/spf13/cobra v1.8.1/go.mod h1:wHxEcudfqmLYa8iTfL+OuZPbBZkmvliBWKIezN3kD9Y=
 github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
 github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
 github.com/sqids/sqids-go v0.4.1 h1:eQKYzmAZbLlRwHeHYPF35QhgxwZHLnlmVj9AkIj/rrw=
 github.com/sqids/sqids-go v0.4.1/go.mod h1:EMwHuPQgSNFS0A49jESTfIQS+066XQTVhukrzEPScl8=
-golang.org/x/sync v0.1.0 h1:wsuoTGHzEhffawBOhz5CYhcrV4IdKZbEyZjBMuTp12o=
-golang.org/x/sync v0.1.0/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.0.0-20210809222454-d867a43fc93e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.14.0 h1:Vz7Qs629MkJkGyHxUlRHizWJRG2j8fbQKjELVSNhy7Q=
-golang.org/x/sys v0.14.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/term v0.6.0 h1:clScbb1cHjoCkyRbWwBEUZ5H/tIFu5TAXIqaZD0Gcjw=
-golang.org/x/term v0.6.0/go.mod h1:m6U89DPEgQRMq3DNkDClhWw02AUbt2daBVO4cn4Hv9U=
-golang.org/x/text v0.3.8 h1:nAL+RVCQ9uMn3vJZbV+MRnydTJFPf8qqY42YiA6MrqY=
-golang.org/x/text v0.3.8/go.mod h1:E6s5w1FMmriuDzIBO73fBruAKo1PCIq6d2Q6DHfQ8WQ=
+golang.org/x/sys v0.25.0 h1:r+8e+loiHxRqhXVl6ML1nO3l1+oFoWbnlu2Ehimmi34=
+golang.org/x/sys v0.25.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/text v0.18.0 h1:XvMDiNzPAl0jr17s6W9lcaIhGUfUORdGCNsuLmPG224=
+golang.org/x/text v0.18.0/go.mod h1:BuEKDfySbSR4drPmRPG/7iBdf8hvFMuRexcpahXilzY=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15 h1:YR8cESwS4TdDjEe65xsg0ogRM/Nc3DYOhEAlW+xobZo=
 gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
-gopkg.in/yaml.v2 v2.2.2 h1:ZCJp+EgiOT7lHqUV2J862kp8Qj64Jo6az82+3Td9dZw=
-gopkg.in/yaml.v2 v2.2.2/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
-gorm.io/driver/sqlite v1.5.4 h1:IqXwXi8M/ZlPzH/947tn5uik3aYQslP9BVveoax0nV0=
-gorm.io/driver/sqlite v1.5.4/go.mod h1:qxAuCol+2r6PannQDpOP1FP6ag3mKi4esLnB/jHed+4=
-gorm.io/gorm v1.25.5 h1:zR9lOiiYf09VNh5Q1gphfyia1JpiClIWG9hQaxB/mls=
-gorm.io/gorm v1.25.5/go.mod h1:hbnx/Oo0ChWMn1BIhpy1oYozzpM15i4YPuHDmfYtwg8=
+gorm.io/driver/sqlite v1.5.6 h1:fO/X46qn5NUEEOZtnjJRWRzZMe8nqJiQ9E+0hi+hKQE=
+gorm.io/driver/sqlite v1.5.6/go.mod h1:U+J8craQU6Fzkcvu8oLeAQmi50TkwPEhHDEjQZXDah4=
+gorm.io/gorm v1.25.12 h1:I0u8i2hWQItBq1WfE0o2+WuL9+8L21K9e2HHSTE/0f8=
+gorm.io/gorm v1.25.12/go.mod h1:xh7N7RHfYlNc5EmcI/El95gXusucDrQnHXe0+CgWcLQ=
--- a/pkg/agents/toolbox/dir_tree.go
+++ b/pkg/agents/toolbox/dir_tree.go
@ -0,0 +1,142 @@
+package toolbox
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+
+	toolutil "git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+)
+
+const TREE_DESCRIPTION = `Retrieve a tree-like view of a directory's contents.
+
+Use these results for your own reference in completing your task, they do not need to be shown to the user.
+
+Example result:
+{
+	"message": "success",
+	"result": ".
+├── a_directory/
+│   ├── file1.txt (100 bytes)
+│   └── file2.txt (200 bytes)
+├── a_file.txt (123 bytes)
+└── another_file.txt (456 bytes)"
+}
+`
+
+var DirTreeTool = api.ToolSpec{
+	Name:        "dir_tree",
+	Description: TREE_DESCRIPTION,
+	Parameters: []api.ToolParameter{
+		{
+			Name:        "relative_path",
+			Type:        "string",
+			Description: "If set, display the tree starting from this path relative to the current one.",
+		},
+		{
+			Name:        "depth",
+			Type:        "integer",
+			Description: "Depth of directory recursion. Defaults to 0 (no recursion), maximum of 5.",
+		},
+	},
+	Impl: func(tool *api.ToolSpec, args map[string]interface{}) (string, error) {
+		var relativeDir string
+		if tmp, ok := args["relative_path"]; ok {
+			relativeDir, ok = tmp.(string)
+			if !ok {
+				return "", fmt.Errorf("expected string for relative_path, got %T", tmp)
+			}
+		}
+
+		var depth int = 0 // Default value if not provided
+		if tmp, ok := args["depth"]; ok {
+			switch v := tmp.(type) {
+			case float64:
+				depth = int(v)
+			case string:
+				var err error
+				if depth, err = strconv.Atoi(v); err != nil {
+					return "", fmt.Errorf("invalid `depth` value, expected integer but got string that cannot convert: %v", tmp)
+				}
+				depth = max(0, min(5, depth))
+			default:
+				return "", fmt.Errorf("expected int or string for max_depth, got %T", tmp)
+			}
+		}
+
+		result := tree(relativeDir, depth)
+		ret, err := result.ToJson()
+		if err != nil {
+			return "", fmt.Errorf("could not serialize result: %v", err)
+		}
+		return ret, nil
+	},
+}
+
+func tree(path string, depth int) api.CallResult {
+	if path == "" {
+		path = "."
+	}
+	ok, reason := toolutil.IsPathWithinCWD(path)
+	if !ok {
+		return api.CallResult{Message: reason}
+	}
+
+	var treeOutput strings.Builder
+	treeOutput.WriteString(path + "\n")
+	err := buildTree(&treeOutput, path, "", depth)
+	if err != nil {
+		return api.CallResult{
+			Message: err.Error(),
+		}
+	}
+
+	return api.CallResult{Result: treeOutput.String()}
+}
+
+func buildTree(output *strings.Builder, path string, prefix string, depth int) error {
+	files, err := os.ReadDir(path)
+	if err != nil {
+		return err
+	}
+
+	for i, file := range files {
+		if strings.HasPrefix(file.Name(), ".") {
+			// Skip hidden files and directories
+			continue
+		}
+
+		isLast := i == len(files)-1
+		var branch string
+		if isLast {
+			branch = "└── "
+		} else {
+			branch = "├── "
+		}
+
+		info, _ := file.Info()
+		size := info.Size()
+		sizeStr := fmt.Sprintf(" (%d bytes)", size)
+
+		output.WriteString(prefix + branch + file.Name())
+		if file.IsDir() {
+			output.WriteString("/\n")
+			if depth > 0 {
+				var nextPrefix string
+				if isLast {
+					nextPrefix = prefix + "    "
+				} else {
+					nextPrefix = prefix + "│   "
+				}
+				buildTree(output, filepath.Join(path, file.Name()), nextPrefix, depth-1)
+			}
+		} else {
+			output.WriteString(sizeStr + "\n")
+		}
+	}
+
+	return nil
+}
--- a/pkg/agents/toolbox/file_insert_lines.go
+++ b/pkg/agents/toolbox/file_insert_lines.go
@ -1,22 +1,22 @@
-package tools
+package toolbox

 import (
 	"fmt"
 	"os"
 	"strings"

-	toolutil "git.mlow.ca/mlow/lmcli/pkg/lmcli/tools/util"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
+	toolutil "git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
 )

 const FILE_INSERT_LINES_DESCRIPTION = `Insert lines into a file, must specify path.

 Make sure your inserts match the flow and indentation of surrounding content.`

-var FileInsertLinesTool = model.Tool{
+var FileInsertLinesTool = api.ToolSpec{
 	Name:        "file_insert_lines",
 	Description: FILE_INSERT_LINES_DESCRIPTION,
-	Parameters: []model.ToolParameter{
+	Parameters: []api.ToolParameter{
 		{
 			Name:        "path",
 			Type:        "string",
@ -36,7 +36,7 @@ var FileInsertLinesTool = model.Tool{
 			Required:    true,
 		},
 	},
-	Impl: func(tool *model.Tool, args map[string]interface{}) (string, error) {
+	Impl: func(tool *api.ToolSpec, args map[string]interface{}) (string, error) {
 		tmp, ok := args["path"]
 		if !ok {
 			return "", fmt.Errorf("path parameter to write_file was not included.")
@ -72,27 +72,27 @@ var FileInsertLinesTool = model.Tool{
 	},
 }

-func fileInsertLines(path string, position int, content string) model.CallResult {
+func fileInsertLines(path string, position int, content string) api.CallResult {
 	ok, reason := toolutil.IsPathWithinCWD(path)
 	if !ok {
-		return model.CallResult{Message: reason}
+		return api.CallResult{Message: reason}
 	}

 	// Read the existing file's content
 	data, err := os.ReadFile(path)
 	if err != nil {
 		if !os.IsNotExist(err) {
-			return model.CallResult{Message: fmt.Sprintf("Could not read path: %s", err.Error())}
+			return api.CallResult{Message: fmt.Sprintf("Could not read path: %s", err.Error())}
 		}
 		_, err = os.Create(path)
 		if err != nil {
-			return model.CallResult{Message: fmt.Sprintf("Could not create new file: %s", err.Error())}
+			return api.CallResult{Message: fmt.Sprintf("Could not create new file: %s", err.Error())}
 		}
 		data = []byte{}
 	}

 	if position < 1 {
-		return model.CallResult{Message: "start_line cannot be less than 1"}
+		return api.CallResult{Message: "start_line cannot be less than 1"}
 	}

 	lines := strings.Split(string(data), "\n")
@ -107,8 +107,8 @@ func fileInsertLines(path string, position int, content string) model.CallResult
 	// Join the lines and write back to the file
 	err = os.WriteFile(path, []byte(newContent), 0644)
 	if err != nil {
-		return model.CallResult{Message: fmt.Sprintf("Could not write to path: %s", err.Error())}
+		return api.CallResult{Message: fmt.Sprintf("Could not write to path: %s", err.Error())}
 	}

-	return model.CallResult{Result: newContent}
+	return api.CallResult{Result: newContent}
 }
--- a/pkg/agents/toolbox/modify_file.go
+++ b/pkg/agents/toolbox/modify_file.go
@ -0,0 +1,178 @@
+package toolbox
+
+import (
+	"fmt"
+	"os"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	toolutil "git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+)
+
+var MODIFY_FILE_DESCRIPTION = []string{
+    "Modify a file. If the file does not exist, it will be created.",
+    "",
+    "Content can be either inserted, replaced, or removed through a combination of the start, stop, and content parameters.",
+    "Use the start and stop line numbers to limit the range of modification to the file.",
+    "If both `start` and `stop` are left unset (or set to 0), the entire file's contents will be updated.",
+    "If `start` is set to n and `stop` to n+1, content will be inserted at line n (the content that was at line n will be shifted below the newly inserted content).",
+    "If only `start` is set, content from the given line and onwards will be updated.",
+    "If only `stop` is set, content up to but not including the given line will be updated.",
+    "",
+    "Examples:",
+    "1. Append to a file:",
+    "   {\"path\": \"example.txt\", \"start\": <last_line_number + 1>, \"content\": \"New content to append\"}",
+    "",
+    "2. Insert at a specific line:",
+    "   {\"path\": \"example.txt\", \"start\": 5, \"stop\": 5, \"content\": \"New line inserted above the previous line 5\"}",
+    "",
+    "3. Replace a range of lines:",
+    "   {\"path\": \"example.txt\", \"start\": 3, \"stop\": 7, \"content\": \"New content replacing lines 3-7\"}",
+    "",
+    "4. Remove a range of lines:",
+    "   {\"path\": \"example.txt\", \"start\": 2, \"stop\": 5}",
+    "",
+    "5. Replace entire file contents:",
+    "   {\"path\": \"example.txt\", \"content\": \"New file contents\"}",
+    "",
+    "6. Update from a specific line to the end of the file:",
+    "   {\"path\": \"example.txt\", \"start\": 10, \"content\": \"New content from line 10 onwards\"}",
+    "",
+    "7. Update from the beginning of the file to a specific line:",
+    "   {\"path\": \"example.txt\", \"stop\": 6, \"content\": \"New content for first 5 lines\"}",
+    "",
+    "Note: Always use specific line numbers based on the current file content. Avoid using arbitrarily large numbers for start or stop.",
+}
+
+var ModifyFile = api.ToolSpec{
+	Name:        "modify_file",
+	Description: strings.Join(MODIFY_FILE_DESCRIPTION, "\n"),
+	Parameters: []api.ToolParameter{
+		{
+			Name:        "path",
+			Type:        "string",
+			Description: "Path of the file to be modified, relative to the current working directory.",
+			Required:    true,
+		},
+		{
+			Name:        "start",
+			Type:        "integer",
+			Description: `Start line of the range to modify (inclusive). If omitted, the beginning of the file is implied.`,
+		},
+		{
+			Name:        "stop",
+			Type:        "integer",
+			Description: `End line of the range to modify (inclusive). If omitted, the end of the file is implied.`,
+		},
+		{
+			Name:        "content",
+			Type:        "string",
+			Description: "Content to insert/replace at the range defined by `start` and `stop`. If omitted, the range is removed.",
+		},
+	},
+	Impl: func(tool *api.ToolSpec, args map[string]interface{}) (string, error) {
+		tmp, ok := args["path"]
+		if !ok {
+			return "", fmt.Errorf("path parameter to modify_file was not included.")
+		}
+		path, ok := tmp.(string)
+		if !ok {
+			return "", fmt.Errorf("Invalid path in function arguments: %v", tmp)
+		}
+		var start int
+		tmp, ok = args["start"]
+		if ok {
+			tmp, ok := tmp.(float64)
+			if !ok {
+				return "", fmt.Errorf("Invalid start in function arguments: %v", tmp)
+			}
+			start = int(tmp)
+		}
+		var stop int
+		tmp, ok = args["stop"]
+		if ok {
+			tmp, ok := tmp.(float64)
+			if !ok {
+				return "", fmt.Errorf("Invalid stop in function arguments: %v", tmp)
+			}
+			stop = int(tmp)
+		}
+		var content string
+		tmp, ok = args["content"]
+		if ok {
+			content, ok = tmp.(string)
+			if !ok {
+				return "", fmt.Errorf("Invalid content in function arguments: %v", tmp)
+			}
+		}
+
+		result := fileModifyContents(path, start, stop, content)
+		ret, err := result.ToJson()
+		if err != nil {
+			return "", fmt.Errorf("Could not serialize result: %v", err)
+		}
+		return ret, nil
+	},
+}
+
+
+func fileModifyContents(path string, startLine int, stopLine int, content string) api.CallResult {
+    ok, reason := toolutil.IsPathWithinCWD(path)
+    if !ok {
+        return api.CallResult{Message: reason}
+    }
+
+    // Read the existing file's content
+    data, err := os.ReadFile(path)
+    if err != nil {
+        if !os.IsNotExist(err) {
+            return api.CallResult{Message: fmt.Sprintf("Could not read path: %s", err.Error())}
+        }
+        _, err = os.Create(path)
+        if err != nil {
+            return api.CallResult{Message: fmt.Sprintf("Could not create new file: %s", err.Error())}
+        }
+        data = []byte{}
+    }
+
+    lines := strings.Split(string(data), "\n")
+    contentLines := strings.Split(strings.TrimSuffix(content, "\n"), "\n")
+
+    // If both start and stop are unset, update the entire file
+    if startLine == 0 && stopLine == 0 {
+        lines = contentLines
+    } else {
+        if startLine < 1 {
+            startLine = 1
+        }
+        if stopLine == 0 || stopLine > len(lines) {
+            stopLine = len(lines)
+        }
+
+        before := lines[:startLine-1]
+        after := lines[stopLine:]
+
+        // Handle insertion case
+        if startLine == stopLine {
+            lines = append(before, append(contentLines, lines[startLine-1:]...)...)
+        } else {
+            // If content is omitted, remove the specified range
+            if content == "" {
+                lines = append(before, after...)
+            } else {
+                lines = append(before, append(contentLines, after...)...)
+            }
+        }
+    }
+
+    newContent := strings.Join(lines, "\n")
+
+    // Write back to the file
+    err = os.WriteFile(path, []byte(newContent), 0644)
+    if err != nil {
+        return api.CallResult{Message: fmt.Sprintf("Could not write to path: %s", err.Error())}
+    }
+
+    return api.CallResult{Result: util.AddLineNumbers(newContent)}
+}
--- a/pkg/agents/toolbox/read_dir.go
+++ b/pkg/agents/toolbox/read_dir.go
@ -1,4 +1,4 @@
-package tools
+package toolbox

 import (
 	"fmt"
@ -6,8 +6,8 @@ import (
 	"path/filepath"
 	"strings"

-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	toolutil "git.mlow.ca/mlow/lmcli/pkg/lmcli/tools/util"
+	toolutil "git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
 )

 const READ_DIR_DESCRIPTION = `Return the contents of the CWD (current working directory).
@ -25,17 +25,17 @@ Example result:
 For files, size represents the size of the file, in bytes.
 For directories, size represents the number of entries in that directory.`

-var ReadDirTool = model.Tool{
+var ReadDirTool = api.ToolSpec{
 	Name:        "read_dir",
 	Description: READ_DIR_DESCRIPTION,
-	Parameters: []model.ToolParameter{
+	Parameters: []api.ToolParameter{
 		{
 			Name:        "relative_dir",
 			Type:        "string",
 			Description: "If set, read the contents of a directory relative to the current one.",
 		},
 	},
-	Impl: func(tool *model.Tool, args map[string]interface{}) (string, error) {
+	Impl: func(tool *api.ToolSpec, args map[string]interface{}) (string, error) {
 		var relativeDir string
 		tmp, ok := args["relative_dir"]
 		if ok {
@ -53,18 +53,18 @@ var ReadDirTool = model.Tool{
 	},
 }

-func readDir(path string) model.CallResult {
+func readDir(path string) api.CallResult {
 	if path == "" {
 		path = "."
 	}
 	ok, reason := toolutil.IsPathWithinCWD(path)
 	if !ok {
-		return model.CallResult{Message: reason}
+		return api.CallResult{Message: reason}
 	}

 	files, err := os.ReadDir(path)
 	if err != nil {
-		return model.CallResult{
+		return api.CallResult{
 			Message: err.Error(),
 		}
 	}
@ -96,5 +96,5 @@ func readDir(path string) model.CallResult {
 		})
 	}

-	return model.CallResult{Result: dirContents}
+	return api.CallResult{Result: dirContents}
 }
--- a/pkg/agents/toolbox/read_file.go
+++ b/pkg/agents/toolbox/read_file.go
@ -1,15 +1,16 @@
-package tools
+package toolbox

 import (
 	"fmt"
 	"os"
-	"strings"

-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	toolutil "git.mlow.ca/mlow/lmcli/pkg/lmcli/tools/util"
+	toolutil "git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
 )

-const READ_FILE_DESCRIPTION = `Read the contents of a text file relative to the current working directory.
+const READ_FILE_DESCRIPTION = `Retrieve the contents of a text file relative to the current working directory.
+
+Use the file contents for your own reference in completing your task, they do not need to be shown to the user.

 Each line of the returned content is prefixed with its line number and a tab (\t).

@ -19,10 +20,10 @@ Example result:
 	"result": "1\tthe contents\n2\tof the file\n"
 }`

-var ReadFileTool = model.Tool{
+var ReadFileTool = api.ToolSpec{
 	Name:        "read_file",
 	Description: READ_FILE_DESCRIPTION,
-	Parameters: []model.ToolParameter{
+	Parameters: []api.ToolParameter{
 		{
 			Name:        "path",
 			Type:        "string",
@ -31,7 +32,7 @@ var ReadFileTool = model.Tool{
 		},
 	},

-	Impl: func(tool *model.Tool, args map[string]interface{}) (string, error) {
+	Impl: func(tool *api.ToolSpec, args map[string]interface{}) (string, error) {
 		tmp, ok := args["path"]
 		if !ok {
 			return "", fmt.Errorf("Path parameter to read_file was not included.")
@ -49,23 +50,16 @@ var ReadFileTool = model.Tool{
 	},
 }

-func readFile(path string) model.CallResult {
+func readFile(path string) api.CallResult {
 	ok, reason := toolutil.IsPathWithinCWD(path)
 	if !ok {
-		return model.CallResult{Message: reason}
+		return api.CallResult{Message: reason}
 	}
 	data, err := os.ReadFile(path)
 	if err != nil {
-		return model.CallResult{Message: fmt.Sprintf("Could not read path: %s", err.Error())}
+		return api.CallResult{Message: fmt.Sprintf("Could not read path: %s", err.Error())}
 	}
-
-	lines := strings.Split(string(data), "\n")
-	content := strings.Builder{}
-	for i, line := range lines {
-		content.WriteString(fmt.Sprintf("%d\t%s\n", i+1, line))
-	}
-
-	return model.CallResult{
-		Result: content.String(),
+	return api.CallResult{
+		Result: toolutil.AddLineNumbers(string(data)),
 	}
 }
--- a/pkg/agents/toolbox/util/util.go
+++ b/pkg/agents/toolbox/util/util.go
@ -65,3 +65,14 @@ func IsPathWithinCWD(path string) (bool, string) {
 	}
 	return true, ""
 }
+
+// AddLineNumbers takes a string of content and returns a new string with line
+// numbers prefixed
+func AddLineNumbers(content string) string {
+	lines := strings.Split(strings.TrimSuffix(content, "\n"), "\n")
+	result := strings.Builder{}
+	for i, line := range lines {
+		result.WriteString(fmt.Sprintf("%d\t%s\n", i+1, line))
+	}
+	return result.String()
+}
--- a/pkg/agents/toolbox/write_file.go
+++ b/pkg/agents/toolbox/write_file.go
@ -1,11 +1,11 @@
-package tools
+package toolbox

 import (
 	"fmt"
 	"os"

-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	toolutil "git.mlow.ca/mlow/lmcli/pkg/lmcli/tools/util"
+	toolutil "git.mlow.ca/mlow/lmcli/pkg/agents/toolbox/util"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
 )

 const WRITE_FILE_DESCRIPTION = `Write the provided contents to a file relative to the current working directory.
@ -15,10 +15,10 @@ Example result:
 	"message": "success"
 }`

-var WriteFileTool = model.Tool{
+var WriteFileTool = api.ToolSpec{
 	Name:        "write_file",
 	Description: WRITE_FILE_DESCRIPTION,
-	Parameters: []model.ToolParameter{
+	Parameters: []api.ToolParameter{
 		{
 			Name:        "path",
 			Type:        "string",
@ -32,7 +32,7 @@ var WriteFileTool = model.Tool{
 			Required:    true,
 		},
 	},
-	Impl: func(t *model.Tool, args map[string]interface{}) (string, error) {
+	Impl: func(t *api.ToolSpec, args map[string]interface{}) (string, error) {
 		tmp, ok := args["path"]
 		if !ok {
 			return "", fmt.Errorf("Path parameter to write_file was not included.")
@ -58,14 +58,14 @@ var WriteFileTool = model.Tool{
 	},
 }

-func writeFile(path string, content string) model.CallResult {
+func writeFile(path string, content string) api.CallResult {
 	ok, reason := toolutil.IsPathWithinCWD(path)
 	if !ok {
-		return model.CallResult{Message: reason}
+		return api.CallResult{Message: reason}
 	}
 	err := os.WriteFile(path, []byte(content), 0644)
 	if err != nil {
-		return model.CallResult{Message: fmt.Sprintf("Could not write to path: %s", err.Error())}
+		return api.CallResult{Message: fmt.Sprintf("Could not write to path: %s", err.Error())}
 	}
-	return model.CallResult{}
+	return api.CallResult{}
 }
--- a/pkg/agents/tools.go
+++ b/pkg/agents/tools.go
@ -0,0 +1,47 @@
+package agents
+
+import (
+	"fmt"
+
+	"git.mlow.ca/mlow/lmcli/pkg/agents/toolbox"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+)
+
+var AvailableTools map[string]api.ToolSpec = map[string]api.ToolSpec{
+	"dir_tree":    toolbox.DirTreeTool,
+	"read_dir":    toolbox.ReadDirTool,
+	"read_file":   toolbox.ReadFileTool,
+	"modify_file": toolbox.ModifyFile,
+	"write_file":  toolbox.WriteFileTool,
+}
+
+func ExecuteToolCalls(calls []api.ToolCall, available []api.ToolSpec) ([]api.ToolResult, error) {
+	var toolResults []api.ToolResult
+	for _, call := range calls {
+		var tool *api.ToolSpec
+		for i := range available {
+			if available[i].Name == call.Name {
+				tool = &available[i]
+				break
+			}
+		}
+		if tool == nil {
+			return nil, fmt.Errorf("Requested tool '%s' is not available. Hallucination?", call.Name)
+		}
+
+		// Execute the tool
+		result, err := tool.Impl(tool, call.Parameters)
+		if err != nil {
+			return nil, fmt.Errorf("Tool '%s' error: %v\n", call.Name, err)
+		}
+
+		toolResult := api.ToolResult{
+			ToolCallID: call.ID,
+			ToolName:   call.Name,
+			Result:     result,
+		}
+
+		toolResults = append(toolResults, toolResult)
+	}
+	return toolResults, nil
+}
--- a/pkg/api/api.go
+++ b/pkg/api/api.go
@ -0,0 +1,126 @@
+package api
+
+import (
+	"encoding/json"
+	"fmt"
+)
+
+type MessageRole string
+
+const (
+	MessageRoleSystem     MessageRole = "system"
+	MessageRoleUser       MessageRole = "user"
+	MessageRoleAssistant  MessageRole = "assistant"
+	MessageRoleToolCall   MessageRole = "tool_call"
+	MessageRoleToolResult MessageRole = "tool_result"
+)
+
+type Message struct {
+	Content     string // TODO: support multi-part messages
+	Role        MessageRole
+	ToolCalls   []ToolCall
+	ToolResults []ToolResult
+}
+
+type ToolSpec struct {
+	Name        string
+	Description string
+	Parameters  []ToolParameter
+	Impl        func(*ToolSpec, map[string]interface{}) (string, error)
+}
+
+type ToolParameter struct {
+	Name        string   `json:"name"`
+	Type        string   `json:"type"` // "string", "integer", "boolean"
+	Required    bool     `json:"required"`
+	Description string   `json:"description"`
+	Enum        []string `json:"enum,omitempty"`
+}
+
+type ToolCall struct {
+	ID         string                 `json:"id" yaml:"-"`
+	Name       string                 `json:"name" yaml:"tool"`
+	Parameters map[string]interface{} `json:"parameters" yaml:"parameters"`
+}
+
+type ToolResult struct {
+	ToolCallID string `json:"toolCallID" yaml:"-"`
+	ToolName   string `json:"toolName,omitempty" yaml:"tool"`
+	Result     string `json:"result,omitempty" yaml:"result"`
+}
+
+func NewMessageWithAssistant(content string) *Message {
+	return &Message{
+		Role:    MessageRoleAssistant,
+		Content: content,
+	}
+}
+
+func NewMessageWithToolCalls(content string, toolCalls []ToolCall) *Message {
+	return &Message{
+		Role:      MessageRoleToolCall,
+		Content:   content,
+		ToolCalls: toolCalls,
+	}
+}
+
+func (m MessageRole) IsAssistant() bool {
+	switch m {
+	case MessageRoleAssistant, MessageRoleToolCall:
+		return true
+	}
+	return false
+}
+
+func (m MessageRole) IsUser() bool {
+	switch m {
+	case MessageRoleUser, MessageRoleToolResult:
+		return true
+	}
+	return false
+}
+
+func (m MessageRole) IsSystem() bool {
+	switch m {
+	case MessageRoleSystem:
+		return true
+	}
+	return false
+}
+
+// FriendlyRole returns a human friendly signifier for the message's role.
+func (m MessageRole) FriendlyRole() string {
+	switch m {
+	case MessageRoleUser:
+		return "You"
+	case MessageRoleSystem:
+		return "System"
+	case MessageRoleAssistant:
+		return "Assistant"
+	case MessageRoleToolCall:
+		return "Tool Call"
+	case MessageRoleToolResult:
+		return "Tool Result"
+	default:
+		return string(m)
+	}
+}
+
+// TODO: remove this
+type CallResult struct {
+	Message string `json:"message"`
+	Result  any    `json:"result,omitempty"`
+}
+
+func (r CallResult) ToJson() (string, error) {
+	if r.Message == "" {
+		// When message not supplied, assume success
+		r.Message = "success"
+	}
+
+	jsonBytes, err := json.Marshal(r)
+	if err != nil {
+		return "", fmt.Errorf("Could not marshal CallResult to JSON: %v\n", err)
+	}
+	return string(jsonBytes), nil
+}
--- a/pkg/cmd/chat.go
+++ b/pkg/cmd/chat.go
@ -3,8 +3,10 @@ package cmd
 import (
 	"fmt"

+	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
 	"git.mlow.ca/mlow/lmcli/pkg/tui"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
 	"github.com/spf13/cobra"
 )

@ -14,12 +16,34 @@ func ChatCmd(ctx *lmcli.Context) *cobra.Command {
 		Short: "Open the chat interface",
 		Long:  `Open the chat interface, optionally on a given conversation.`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			// TODO: implement jump-to-conversation logic
-			shortname := ""
-			if len(args) == 1 {
-				shortname = args[0]
+			err := validateGenerationFlags(ctx, cmd)
+			if err != nil {
+				return err
 			}
-			err := tui.Launch(ctx, shortname)
+
+			var opts []tui.LaunchOption
+
+			list, err := cmd.Flags().GetBool("list")
+			if err != nil {
+				return err
+			}
+
+			if !list && len(args) == 1 {
+				shortname := args[0]
+				if shortname != ""{
+					conv, err := cmdutil.LookupConversationE(ctx, shortname)
+					if err != nil {
+						return err
+					}
+					opts = append(opts, tui.WithInitialConversation(conv))
+				}
+			}
+
+			if list {
+				opts = append(opts, tui.WithInitialView(shared.ViewConversations))
+			}
+
+			err = tui.Launch(ctx, opts...)
 			if err != nil {
 				return fmt.Errorf("Error fetching LLM response: %v", err)
 			}
@ -30,8 +54,13 @@ func ChatCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}
+
+	// -l, --list
+	cmd.Flags().BoolP("list", "l", false, "View/manage conversations")
+
+	applyGenerationFlags(ctx, cmd)
 	return cmd
 }
--- a/pkg/cmd/clone.go
+++ b/pkg/cmd/clone.go
@ -5,7 +5,6 @@ import (

 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -28,36 +27,12 @@ func CloneCmd(ctx *lmcli.Context) *cobra.Command {
 				return err
 			}

-			messagesToCopy, err := ctx.Store.Messages(toClone)
+			clone, messageCnt, err := ctx.Conversations.CloneConversation(*toClone)
 			if err != nil {
-				return fmt.Errorf("Could not retrieve messages for conversation: %s", toClone.ShortName.String)
+				return fmt.Errorf("Failed to clone conversation: %v", err)
 			}

-			clone := &model.Conversation{
-				Title: toClone.Title + " - Clone",
-			}
-			if err := ctx.Store.SaveConversation(clone); err != nil {
-				return fmt.Errorf("Cloud not create clone: %s", err)
-			}
-
-			var errors []error
-			messageCnt := 0
-			for _, message := range messagesToCopy {
-				newMessage := message
-				newMessage.ConversationID = clone.ID
-				newMessage.ID = 0
-				if err := ctx.Store.SaveMessage(&newMessage); err != nil {
-					errors = append(errors, err)
-				} else {
-					messageCnt++
-				}
-			}
-
-			if len(errors) > 0 {
-				return fmt.Errorf("Messages failed to be cloned: %v", errors)
-			}
-
-			fmt.Printf("Cloned %d messages to: %s\n", messageCnt, clone.Title)
+			fmt.Printf("Cloned %d messages to: %s - %s\n", messageCnt, clone.ShortName.String, clone.Title)
 			return nil
 		},
 		ValidArgsFunction: func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
@ -65,7 +40,7 @@ func CloneCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}
 	return cmd
--- a/pkg/cmd/cmd.go
+++ b/pkg/cmd/cmd.go
@ -1,6 +1,8 @@
 package cmd

 import (
+	"fmt"
+	"slices"
 	"strings"

 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
@ -8,10 +10,6 @@ import (
 	"github.com/spf13/cobra"
 )

-var (
-	systemPromptFile string
-)
-
 func RootCmd(ctx *lmcli.Context) *cobra.Command {
 	var root = &cobra.Command{
 		Use:           "lmcli <command> [flags]",
@ -23,58 +21,72 @@ func RootCmd(ctx *lmcli.Context) *cobra.Command {
 		},
 	}

-	chatCmd := ChatCmd(ctx)
-	continueCmd := ContinueCmd(ctx)
-	cloneCmd := CloneCmd(ctx)
-	editCmd := EditCmd(ctx)
-	listCmd := ListCmd(ctx)
-	newCmd := NewCmd(ctx)
-	promptCmd := PromptCmd(ctx)
-	renameCmd := RenameCmd(ctx)
-	replyCmd := ReplyCmd(ctx)
-	retryCmd := RetryCmd(ctx)
-	rmCmd := RemoveCmd(ctx)
-	viewCmd := ViewCmd(ctx)
-
-	inputCmds := []*cobra.Command{newCmd, promptCmd, replyCmd, retryCmd, continueCmd, editCmd}
-	for _, cmd := range inputCmds {
-		cmd.Flags().StringVar(ctx.Config.Defaults.Model, "model", *ctx.Config.Defaults.Model, "Which model to use")
-		cmd.RegisterFlagCompletionFunc("model", func(*cobra.Command, []string, string) ([]string, cobra.ShellCompDirective) {
-			return ctx.GetModels(), cobra.ShellCompDirectiveDefault
-		})
-		cmd.Flags().IntVar(ctx.Config.Defaults.MaxTokens, "length", *ctx.Config.Defaults.MaxTokens, "Maximum response tokens")
-		cmd.Flags().StringVar(ctx.Config.Defaults.SystemPrompt, "system-prompt", *ctx.Config.Defaults.SystemPrompt, "System prompt")
-		cmd.Flags().StringVar(&systemPromptFile, "system-prompt-file", "", "A path to a file containing the system prompt")
-		cmd.MarkFlagsMutuallyExclusive("system-prompt", "system-prompt-file")
-	}
-
 	root.AddCommand(
-		chatCmd,
-		cloneCmd,
-		continueCmd,
-		editCmd,
-		listCmd,
-		newCmd,
-		promptCmd,
-		renameCmd,
-		replyCmd,
-		retryCmd,
-		rmCmd,
-		viewCmd,
+		ChatCmd(ctx),
+		ContinueCmd(ctx),
+		CloneCmd(ctx),
+		EditCmd(ctx),
+		ListCmd(ctx),
+		NewCmd(ctx),
+		PromptCmd(ctx),
+		RenameCmd(ctx),
+		ReplyCmd(ctx),
+		RetryCmd(ctx),
+		RemoveCmd(ctx),
+		ViewCmd(ctx),
 	)

 	return root
 }

-func getSystemPrompt(ctx *lmcli.Context) string {
-	if systemPromptFile != "" {
-		content, err := util.ReadFileContents(systemPromptFile)
-		if err != nil {
-			lmcli.Fatal("Could not read file contents at %s: %v\n", systemPromptFile, err)
-		}
-		return content
+func applyGenerationFlags(ctx *lmcli.Context, cmd *cobra.Command) {
+	f := cmd.Flags()
+
+	// -m, --model
+	f.StringVarP(
+		ctx.Config.Defaults.Model, "model", "m",
+		*ctx.Config.Defaults.Model, "Which model to generate a response with",
+	)
+	cmd.RegisterFlagCompletionFunc("model", func(*cobra.Command, []string, string) ([]string, cobra.ShellCompDirective) {
+		return ctx.GetModels(), cobra.ShellCompDirectiveDefault
+	})
+
+	// -a, --agent
+	f.StringVarP(&ctx.Config.Defaults.Agent, "agent", "a", ctx.Config.Defaults.Agent, "Which agent to interact with")
+	cmd.RegisterFlagCompletionFunc("agent", func(*cobra.Command, []string, string) ([]string, cobra.ShellCompDirective) {
+		return ctx.GetAgents(), cobra.ShellCompDirectiveDefault
+	})
+
+	// --max-length
+	f.IntVar(ctx.Config.Defaults.MaxTokens, "max-length", *ctx.Config.Defaults.MaxTokens, "Maximum response tokens")
+	// --temperature
+	f.Float32VarP(ctx.Config.Defaults.Temperature, "temperature", "t", *ctx.Config.Defaults.Temperature, "Sampling temperature")
+
+	// --system-prompt
+	f.StringVar(&ctx.Config.Defaults.SystemPrompt, "system-prompt", ctx.Config.Defaults.SystemPrompt, "System prompt")
+	// --system-prompt-file
+	f.StringVar(&ctx.Config.Defaults.SystemPromptFile, "system-prompt-file", ctx.Config.Defaults.SystemPromptFile, "A path to a file containing the system prompt")
+	cmd.MarkFlagsMutuallyExclusive("system-prompt", "system-prompt-file")
+}
+
+func validateGenerationFlags(ctx *lmcli.Context, cmd *cobra.Command) error {
+	f := cmd.Flags()
+	model, err := f.GetString("model")
+	if err != nil {
+		return fmt.Errorf("Error parsing --model: %w", err)
 	}
-	return *ctx.Config.Defaults.SystemPrompt
+	if model != "" && !slices.Contains(ctx.GetModels(), model) {
+		return fmt.Errorf("Unknown model: %s", model)
+	}
+
+	agent, err := f.GetString("agent")
+	if err != nil {
+		return fmt.Errorf("Error parsing --agent: %w", err)
+	}
+	if agent != "" && agent != "none"  && !slices.Contains(ctx.GetAgents(), agent) {
+		return fmt.Errorf("Unknown agent: %s", agent)
+	}
+	return nil
 }

 // inputFromArgsOrEditor returns either the provided input from the args slice
--- a/pkg/cmd/continue.go
+++ b/pkg/cmd/continue.go
@ -4,9 +4,9 @@ import (
 	"fmt"
 	"strings"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -23,10 +23,15 @@ func ContinueCmd(ctx *lmcli.Context) *cobra.Command {
 			return nil
 		},
 		RunE: func(cmd *cobra.Command, args []string) error {
-			shortName := args[0]
-			conversation := cmdutil.LookupConversation(ctx, shortName)
+			err := validateGenerationFlags(ctx, cmd)
+			if err != nil {
+				return err
+			}

-			messages, err := ctx.Store.Messages(conversation)
+			shortName := args[0]
+			c := cmdutil.LookupConversation(ctx, shortName)
+
+			messages, err := ctx.Conversations.PathToLeaf(c.SelectedRoot)
 			if err != nil {
 				return fmt.Errorf("could not retrieve conversation messages: %v", err)
 			}
@ -36,7 +41,7 @@ func ContinueCmd(ctx *lmcli.Context) *cobra.Command {
 			}

 			lastMessage := &messages[len(messages)-1]
-			if lastMessage.Role != model.MessageRoleAssistant {
+			if lastMessage.Role != api.MessageRoleAssistant {
 				return fmt.Errorf("the last message in the conversation is not an assistant message")
 			}

@ -44,16 +49,16 @@ func ContinueCmd(ctx *lmcli.Context) *cobra.Command {
 			fmt.Print(lastMessage.Content)

 			// Submit the LLM request, allowing it to continue the last message
-			continuedOutput, err := cmdutil.FetchAndShowCompletion(ctx, messages, nil)
+			continuedOutput, err := cmdutil.Prompt(ctx, messages, nil)
 			if err != nil {
 				return fmt.Errorf("error fetching LLM response: %v", err)
 			}

 			// Append the new response to the original message
-			lastMessage.Content += strings.TrimRight(continuedOutput, "\n\t ")
+			lastMessage.Content += strings.TrimRight(continuedOutput.Content, "\n\t ")

 			// Update the original message
-			err = ctx.Store.UpdateMessage(lastMessage)
+			err = ctx.Conversations.UpdateMessage(lastMessage)
 			if err != nil {
 				return fmt.Errorf("could not update the last message: %v", err)
 			}
@ -65,8 +70,9 @@ func ContinueCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}
+	applyGenerationFlags(ctx, cmd)
 	return cmd
 }
--- a/pkg/cmd/edit.go
+++ b/pkg/cmd/edit.go
@ -3,9 +3,9 @@ package cmd
 import (
 	"fmt"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -22,11 +22,11 @@ func EditCmd(ctx *lmcli.Context) *cobra.Command {
 		},
 		RunE: func(cmd *cobra.Command, args []string) error {
 			shortName := args[0]
-			conversation := cmdutil.LookupConversation(ctx, shortName)
+			c := cmdutil.LookupConversation(ctx, shortName)

-			messages, err := ctx.Store.Messages(conversation)
+			messages, err := ctx.Conversations.PathToLeaf(c.SelectedRoot)
 			if err != nil {
-				return fmt.Errorf("Could not retrieve messages for conversation: %s", conversation.Title)
+				return fmt.Errorf("Could not retrieve messages for conversation: %s", c.Title)
 			}

 			offset, _ := cmd.Flags().GetInt("offset")
@ -39,21 +39,7 @@ func EditCmd(ctx *lmcli.Context) *cobra.Command {
 			}

 			desiredIdx := len(messages) - 1 - offset
-
-			// walk backwards through the conversation deleting messages until and
-			// including the last user message
-			toRemove := []model.Message{}
-			var toEdit *model.Message
-			for i := len(messages) - 1; i >= 0; i-- {
-				if i == desiredIdx {
-					toEdit = &messages[i]
-				}
-				toRemove = append(toRemove, messages[i])
-				messages = messages[:i]
-				if toEdit != nil {
-					break
-				}
-			}
+			toEdit := messages[desiredIdx]

 			newContents := inputFromArgsOrEditor(args[1:], "# Save when finished editing\n", toEdit.Content)
 			switch newContents {
@ -63,38 +49,51 @@ func EditCmd(ctx *lmcli.Context) *cobra.Command {
 				return fmt.Errorf("No message was provided.")
 			}

+			toEdit.Content = newContents
+
 			role, _ := cmd.Flags().GetString("role")
-			if role == "" {
-				role = string(toEdit.Role)
-			} else if role != string(model.MessageRoleUser) && role != string(model.MessageRoleAssistant) {
-				return fmt.Errorf("Invalid role specified. Please use 'user' or 'assistant'.")
-			}
-
-			for _, message := range toRemove {
-				err = ctx.Store.DeleteMessage(&message)
-				if err != nil {
-					lmcli.Warn("Could not delete message: %v\n", err)
+			if role != "" {
+				if role != string(api.MessageRoleUser) && role != string(api.MessageRoleAssistant) {
+					return fmt.Errorf("Invalid role specified. Please use 'user' or 'assistant'.")
 				}
+				toEdit.Role = api.MessageRole(role)
 			}

-			cmdutil.HandleConversationReply(ctx, conversation, true, model.Message{
-				ConversationID:  conversation.ID,
-				Role:            model.MessageRole(role),
-				Content: newContents,
-			})
-			return nil
+			// Update the message in-place
+			inplace, _ := cmd.Flags().GetBool("in-place")
+			if inplace {
+				return ctx.Conversations.UpdateMessage(&toEdit)
+			}
+
+			// Otherwise, create a branch for the edited message
+			message, _, err := ctx.Conversations.CloneBranch(toEdit)
+			if err != nil {
+				return err
+			}
+
+			if desiredIdx > 0 {
+				// update selected reply
+				messages[desiredIdx-1].SelectedReply = message
+				err = ctx.Conversations.UpdateMessage(&messages[desiredIdx-1])
+			} else {
+				// update selected root
+				c.SelectedRoot = message
+				err = ctx.Conversations.UpdateConversation(c)
+			}
+			return err
 		},
 		ValidArgsFunction: func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
 			compMode := cobra.ShellCompDirectiveNoFileComp
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}

+	cmd.Flags().BoolP("in-place", "i", true, "Edit the message in-place, rather than creating a branch")
 	cmd.Flags().Int("offset", 1, "Offset from the last message to edit")
-	cmd.Flags().StringP("role", "r", "", "Role of the edited message (user or assistant)")
+	cmd.Flags().StringP("role", "r", "", "Change the role of the edited message (user or assistant)")

 	return cmd
 }
--- a/pkg/cmd/list.go
+++ b/pkg/cmd/list.go
@ -2,7 +2,6 @@ package cmd

 import (
 	"fmt"
-	"slices"
 	"time"

 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
@ -21,9 +20,9 @@ func ListCmd(ctx *lmcli.Context) *cobra.Command {
 		Short:   "List conversations",
 		Long:    `List conversations in order of recent activity`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			conversations, err := ctx.Store.Conversations()
+			list, err := ctx.Conversations.LoadConversationList()
 			if err != nil {
-				return fmt.Errorf("Could not fetch conversations: %v", err)
+				return fmt.Errorf("Could not load conversations: %v", err)
 			}

 			type Category struct {
@ -58,17 +57,12 @@ func ListCmd(ctx *lmcli.Context) *cobra.Command {

 			all, _ := cmd.Flags().GetBool("all")

-			for _, conversation := range conversations {
-				lastMessage, err := ctx.Store.LastMessage(&conversation)
-				if lastMessage == nil || err != nil {
-					continue
-				}
-
-				messageAge := now.Sub(lastMessage.CreatedAt)
+			for _, item := range list.Items {
+				age := now.Sub(item.LastMessageAt)

 				var category string
 				for _, c := range categories {
-					if messageAge < c.cutoff {
+					if age < c.cutoff {
 						category = c.name
 						break
 					}
@ -76,14 +70,14 @@ func ListCmd(ctx *lmcli.Context) *cobra.Command {

 				formatted := fmt.Sprintf(
 					"%s - %s - %s",
-					conversation.ShortName.String,
-					util.HumanTimeElapsedSince(messageAge),
-					conversation.Title,
+					item.ShortName,
+					util.HumanTimeElapsedSince(age),
+					item.Title,
 				)

 				categorized[category] = append(
 					categorized[category],
-					ConversationLine{messageAge, formatted},
+					ConversationLine{age, formatted},
 				)
 			}

@ -96,14 +90,10 @@ func ListCmd(ctx *lmcli.Context) *cobra.Command {
 					continue
 				}

-				slices.SortFunc(conversationLines, func(a, b ConversationLine) int {
-					return int(a.timeSinceReply - b.timeSinceReply)
-				})
-
 				fmt.Printf("%s:\n", category.name)
 				for _, conv := range conversationLines {
 					if conversationsPrinted >= count && !all {
-						fmt.Printf("%d remaining message(s), use --all to view.\n", len(conversations)-conversationsPrinted)
+						fmt.Printf("%d remaining conversation(s), use --all to view.\n", list.Total-conversationsPrinted)
 						break outer
 					}

@ -115,8 +105,8 @@ func ListCmd(ctx *lmcli.Context) *cobra.Command {
 		},
 	}

-	cmd.Flags().Bool("all", false, "Show all conversations")
-	cmd.Flags().Int("count", LS_COUNT, "How many conversations to show")
+	cmd.Flags().BoolP("all", "a", false, "Show all conversations")
+	cmd.Flags().IntP("count", "c", LS_COUNT, "How many conversations to show")

 	return cmd
 }
--- a/pkg/cmd/new.go
+++ b/pkg/cmd/new.go
@ -3,9 +3,10 @@ package cmd
 import (
 	"fmt"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -15,46 +16,42 @@ func NewCmd(ctx *lmcli.Context) *cobra.Command {
 		Short: "Start a new conversation",
 		Long:  `Start a new conversation with the Large Language Model.`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			messageContents := inputFromArgsOrEditor(args, "# What would you like to say?\n", "")
-			if messageContents == "" {
+			err := validateGenerationFlags(ctx, cmd)
+			if err != nil {
+				return err
+			}
+
+			input := inputFromArgsOrEditor(args, "# Start a new conversation below\n", "")
+			if input == "" {
 				return fmt.Errorf("No message was provided.")
 			}

-			conversation := &model.Conversation{}
-			err := ctx.Store.SaveConversation(conversation)
+			messages := []conversation.Message{{
+				Role:    api.MessageRoleUser,
+				Content: input,
+			}}
+
+			conversation, messages, err := ctx.Conversations.StartConversation(messages...)
 			if err != nil {
-				return fmt.Errorf("Could not save new conversation: %v", err)
+				return fmt.Errorf("Could not start a new conversation: %v", err)
 			}

-			messages := []model.Message{
-				{
-					ConversationID:  conversation.ID,
-					Role:            model.MessageRoleSystem,
-					Content: getSystemPrompt(ctx),
-				},
-				{
-					ConversationID:  conversation.ID,
-					Role:            model.MessageRoleUser,
-					Content: messageContents,
-				},
-			}
+			cmdutil.HandleReply(ctx, &messages[len(messages)-1], true)

-			cmdutil.HandleConversationReply(ctx, conversation, true, messages...)
-
-			title, err := cmdutil.GenerateTitle(ctx, conversation)
+			title, err := cmdutil.GenerateTitle(ctx, messages)
 			if err != nil {
-				lmcli.Warn("Could not generate title for conversation: %v\n", err)
+				lmcli.Warn("Could not generate title for conversation %s: %v\n", conversation.ShortName.String, err)
 			}

 			conversation.Title = title
-
-			err = ctx.Store.SaveConversation(conversation)
+			err = ctx.Conversations.UpdateConversation(conversation)
 			if err != nil {
-				lmcli.Warn("Could not save conversation after generating title: %v\n", err)
+				lmcli.Warn("Could not save conversation title: %v\n", err)
 			}
 			return nil
 		},
 	}

+	applyGenerationFlags(ctx, cmd)
 	return cmd
 }
--- a/pkg/cmd/prompt.go
+++ b/pkg/cmd/prompt.go
@ -3,9 +3,10 @@ package cmd
 import (
 	"fmt"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -15,28 +16,29 @@ func PromptCmd(ctx *lmcli.Context) *cobra.Command {
 		Short: "Do a one-shot prompt",
 		Long:  `Prompt the Large Language Model and get a response.`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			message := inputFromArgsOrEditor(args, "# What would you like to say?\n", "")
-			if message == "" {
+			err := validateGenerationFlags(ctx, cmd)
+			if err != nil {
+				return err
+			}
+
+			input := inputFromArgsOrEditor(args, "# Write your prompt below\n", "")
+			if input == "" {
 				return fmt.Errorf("No message was provided.")
 			}

-			messages := []model.Message{
-				{
-					Role:            model.MessageRoleSystem,
-					Content: getSystemPrompt(ctx),
-				},
-				{
-					Role:            model.MessageRoleUser,
-					Content: message,
-				},
-			}
+			messages := []conversation.Message{{
+				Role:    api.MessageRoleUser,
+				Content: input,
+			}}

-			_, err := cmdutil.FetchAndShowCompletion(ctx, messages, nil)
+			_, err = cmdutil.Prompt(ctx, messages, nil)
 			if err != nil {
 				return fmt.Errorf("Error fetching LLM response: %v", err)
 			}
 			return nil
 		},
 	}
+
+	applyGenerationFlags(ctx, cmd)
 	return cmd
 }
--- a/pkg/cmd/remove.go
+++ b/pkg/cmd/remove.go
@ -5,8 +5,8 @@ import (
 	"strings"

 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -23,14 +23,14 @@ func RemoveCmd(ctx *lmcli.Context) *cobra.Command {
 			return nil
 		},
 		RunE: func(cmd *cobra.Command, args []string) error {
-			var toRemove []*model.Conversation
+			var toRemove []*conversation.Conversation
 			for _, shortName := range args {
 				conversation := cmdutil.LookupConversation(ctx, shortName)
 				toRemove = append(toRemove, conversation)
 			}
 			var errors []error
 			for _, c := range toRemove {
-				err := ctx.Store.DeleteConversation(c)
+				err := ctx.Conversations.DeleteConversation(c)
 				if err != nil {
 					errors = append(errors, fmt.Errorf("Could not remove conversation %s: %v", c.ShortName.String, err))
 				}
@ -44,7 +44,7 @@ func RemoveCmd(ctx *lmcli.Context) *cobra.Command {
 			compMode := cobra.ShellCompDirectiveNoFileComp
 			var completions []string
 		outer:
-			for _, completion := range ctx.Store.ConversationShortNameCompletions(toComplete) {
+			for _, completion := range ctx.Conversations.ConversationShortNameCompletions(toComplete) {
 				parts := strings.Split(completion, "\t")
 				for _, arg := range args {
 					if parts[0] == arg {
--- a/pkg/cmd/rename.go
+++ b/pkg/cmd/rename.go
@ -24,12 +24,17 @@ func RenameCmd(ctx *lmcli.Context) *cobra.Command {
 		RunE: func(cmd *cobra.Command, args []string) error {
 			shortName := args[0]
 			conversation := cmdutil.LookupConversation(ctx, shortName)
+
 			var err error
+			var title string

 			generate, _ := cmd.Flags().GetBool("generate")
-			var title string
 			if generate {
-				title, err = cmdutil.GenerateTitle(ctx, conversation)
+				messages, err := ctx.Conversations.PathToLeaf(conversation.SelectedRoot)
+				if err != nil {
+					return fmt.Errorf("Could not retrieve conversation messages: %v", err)
+				}
+				title, err = cmdutil.GenerateTitle(ctx, messages)
 				if err != nil {
 					return fmt.Errorf("Could not generate conversation title: %v", err)
 				}
@ -41,9 +46,9 @@ func RenameCmd(ctx *lmcli.Context) *cobra.Command {
 			}

 			conversation.Title = title
-			err = ctx.Store.SaveConversation(conversation)
+			err = ctx.Conversations.UpdateConversation(conversation)
 			if err != nil {
-				lmcli.Warn("Could not save conversation with new title: %v\n", err)
+				lmcli.Warn("Could not update conversation title: %v\n", err)
 			}
 			return nil
 		},
@ -52,7 +57,7 @@ func RenameCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}

--- a/pkg/cmd/reply.go
+++ b/pkg/cmd/reply.go
@ -3,9 +3,10 @@ package cmd
 import (
 	"fmt"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -22,17 +23,21 @@ func ReplyCmd(ctx *lmcli.Context) *cobra.Command {
 			return nil
 		},
 		RunE: func(cmd *cobra.Command, args []string) error {
+			err := validateGenerationFlags(ctx, cmd)
+			if err != nil {
+				return err
+			}
+
 			shortName := args[0]
-			conversation := cmdutil.LookupConversation(ctx, shortName)
+			c := cmdutil.LookupConversation(ctx, shortName)

 			reply := inputFromArgsOrEditor(args[1:], "# How would you like to reply?\n", "")
 			if reply == "" {
 				return fmt.Errorf("No reply was provided.")
 			}

-			cmdutil.HandleConversationReply(ctx, conversation, true, model.Message{
-				ConversationID:  conversation.ID,
-				Role:            model.MessageRoleUser,
+			cmdutil.HandleConversationReply(ctx, c, true, conversation.Message{
+				Role:    api.MessageRoleUser,
 				Content: reply,
 			})
 			return nil
@ -42,8 +47,10 @@ func ReplyCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}
+
+	applyGenerationFlags(ctx, cmd)
 	return cmd
 }
--- a/pkg/cmd/retry.go
+++ b/pkg/cmd/retry.go
@ -3,9 +3,9 @@ package cmd
 import (
 	"fmt"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
 	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"github.com/spf13/cobra"
 )

@ -13,7 +13,7 @@ func RetryCmd(ctx *lmcli.Context) *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "retry <conversation>",
 		Short: "Retry the last user reply in a conversation",
-		Long:  `Re-prompt the conversation up to the last user response. Can be used to regenerate the last assistant reply, or simply generate one if an error occurred.`,
+		Long:  `Prompt the conversation from the last user response.`,
 		Args: func(cmd *cobra.Command, args []string) error {
 			argCount := 1
 			if err := cobra.MinimumNArgs(argCount)(cmd, args); err != nil {
@ -22,28 +22,44 @@ func RetryCmd(ctx *lmcli.Context) *cobra.Command {
 			return nil
 		},
 		RunE: func(cmd *cobra.Command, args []string) error {
-			shortName := args[0]
-			conversation := cmdutil.LookupConversation(ctx, shortName)
-
-			messages, err := ctx.Store.Messages(conversation)
+			err := validateGenerationFlags(ctx, cmd)
 			if err != nil {
-				return fmt.Errorf("Could not retrieve messages for conversation: %s", conversation.Title)
+				return err
 			}

-			// walk backwards through the conversation and delete messages, break
-			// when we find the latest user response
-			for i := len(messages) - 1; i >= 0; i-- {
-				if messages[i].Role == model.MessageRoleUser {
-					break
-				}
+			shortName := args[0]
+			c := cmdutil.LookupConversation(ctx, shortName)

-				err = ctx.Store.DeleteMessage(&messages[i])
-				if err != nil {
-					lmcli.Warn("Could not delete previous reply: %v\n", err)
-				}
+			// Load the complete thread from the root message
+			messages, err := ctx.Conversations.PathToLeaf(c.SelectedRoot)
+			if err != nil {
+				return fmt.Errorf("Could not retrieve messages for conversation: %s", c.Title)
 			}

-			cmdutil.HandleConversationReply(ctx, conversation, true)
+			offset, _ := cmd.Flags().GetInt("offset")
+			if offset < 0 {
+				offset = -offset
+			}
+
+			if offset > len(messages)-1 {
+				return fmt.Errorf("Offset %d is before the start of the conversation.", offset)
+			}
+
+			retryFromIdx := len(messages) - 1 - offset
+
+			// decrease retryFromIdx until we hit a user message
+			for retryFromIdx >= 0 && messages[retryFromIdx].Role != api.MessageRoleUser {
+				retryFromIdx--
+			}
+
+			if messages[retryFromIdx].Role != api.MessageRoleUser {
+				return fmt.Errorf("No user messages to retry")
+			}
+
+			fmt.Printf("Idx: %d Message: %v\n", retryFromIdx, messages[retryFromIdx])
+
+			// Start a new branch at the last user message
+			cmdutil.HandleReply(ctx, &messages[retryFromIdx], true)
 			return nil
 		},
 		ValidArgsFunction: func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
@ -51,8 +67,12 @@ func RetryCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}
+
+	cmd.Flags().Int("offset", 0, "Offset from the last message to retry from.")
+
+	applyGenerationFlags(ctx, cmd)
 	return cmd
 }
--- a/pkg/cmd/util/util.go
+++ b/pkg/cmd/util/util.go
@ -2,42 +2,59 @@ package util

 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"os"
 	"strings"
 	"time"

+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
 	"git.mlow.ca/mlow/lmcli/pkg/util"
 	"github.com/charmbracelet/lipgloss"
 )

-// fetchAndShowCompletion prompts the LLM with the given messages and streams
-// the response to stdout. Returns all model reply messages.
-func FetchAndShowCompletion(ctx *lmcli.Context, messages []model.Message, callback func(model.Message)) (string, error) {
-	content := make(chan string) // receives the reponse from LLM
-	defer close(content)
-
-	// render all content received over the channel
-	go ShowDelayedContent(content)
-
-	completionProvider, err := ctx.GetCompletionProvider(*ctx.Config.Defaults.Model)
+// Prompt prompts the configured the configured model and streams the response
+// to stdout. Returns all model reply messages.
+func Prompt(ctx *lmcli.Context, messages []conversation.Message, callback func(conversation.Message)) (*api.Message, error) {
+	m, _, p, err := ctx.GetModelProvider(*ctx.Config.Defaults.Model, "")
 	if err != nil {
-		return "", err
+		return nil, err
 	}

-	requestParams := model.RequestParameters{
-		Model:       *ctx.Config.Defaults.Model,
+	params := provider.RequestParameters{
+		Model:       m,
 		MaxTokens:   *ctx.Config.Defaults.MaxTokens,
 		Temperature: *ctx.Config.Defaults.Temperature,
-		ToolBag:     ctx.EnabledTools,
 	}

-	response, err := completionProvider.CreateChatCompletionStream(
-		context.Background(), requestParams, messages, callback, content,
+	system := ctx.DefaultSystemPrompt()
+
+	agent := ctx.GetAgent(ctx.Config.Defaults.Agent)
+	if agent != nil {
+		if agent.SystemPrompt != "" {
+			system = agent.SystemPrompt
+		}
+		params.Toolbox = agent.Toolbox
+	}
+
+	if system != "" {
+		messages = conversation.ApplySystemPrompt(messages, system, false)
+	}
+
+	content := make(chan provider.Chunk)
+	defer close(content)
+
+	// render the content received over the channel
+	go ShowDelayedContent(content)
+
+	reply, err := p.CreateChatCompletionStream(
+		context.Background(), params, conversation.MessagesToAPI(messages), content,
 	)
-	if response != "" {
+
+	if reply.Content != "" {
 		// there was some content, so break to a new line after it
 		fmt.Println()

@ -46,85 +63,99 @@ func FetchAndShowCompletion(ctx *lmcli.Context, messages []model.Message, callba
 			err = nil
 		}
 	}
-	return response, nil
+	return reply, err
 }

 // lookupConversation either returns the conversation found by the
 // short name or exits the program
-func LookupConversation(ctx *lmcli.Context, shortName string) *model.Conversation {
-	c, err := ctx.Store.ConversationByShortName(shortName)
+func LookupConversation(ctx *lmcli.Context, shortName string) *conversation.Conversation {
+	c, err := ctx.Conversations.FindConversationByShortName(shortName)
 	if err != nil {
 		lmcli.Fatal("Could not lookup conversation: %v\n", err)
 	}
 	if c.ID == 0 {
-		lmcli.Fatal("Conversation not found with short name: %s\n", shortName)
+		lmcli.Fatal("Conversation not found: %s\n", shortName)
 	}
 	return c
 }

-func LookupConversationE(ctx *lmcli.Context, shortName string) (*model.Conversation, error) {
-	c, err := ctx.Store.ConversationByShortName(shortName)
+func LookupConversationE(ctx *lmcli.Context, shortName string) (*conversation.Conversation, error) {
+	c, err := ctx.Conversations.FindConversationByShortName(shortName)
 	if err != nil {
 		return nil, fmt.Errorf("Could not lookup conversation: %v", err)
 	}
 	if c.ID == 0 {
-		return nil, fmt.Errorf("Conversation not found with short name: %s", shortName)
+		return nil, fmt.Errorf("Conversation not found: %s", shortName)
 	}
 	return c, nil
 }

+func HandleConversationReply(ctx *lmcli.Context, c *conversation.Conversation, persist bool, toSend ...conversation.Message) {
+	messages, err := ctx.Conversations.PathToLeaf(c.SelectedRoot)
+	if err != nil {
+		lmcli.Fatal("Could not load messages: %v\n", err)
+	}
+	HandleReply(ctx, &messages[len(messages)-1], persist, toSend...)
+}
+
 // handleConversationReply handles sending messages to an existing
 // conversation, optionally persisting both the sent replies and responses.
-func HandleConversationReply(ctx *lmcli.Context, c *model.Conversation, persist bool, toSend ...model.Message) {
-	existing, err := ctx.Store.Messages(c)
-	if err != nil {
-		lmcli.Fatal("Could not retrieve messages for conversation: %s\n", c.Title)
+func HandleReply(ctx *lmcli.Context, to *conversation.Message, persist bool, messages ...conversation.Message) {
+	if to == nil {
+		lmcli.Fatal("Can't prompt from an empty message.")
 	}

-	if persist {
-		for _, message := range toSend {
-			err = ctx.Store.SaveMessage(&message)
-			if err != nil {
-				lmcli.Warn("Could not save %s message: %v\n", message.Role, err)
-			}
+	existing, err := ctx.Conversations.PathToRoot(to)
+	if err != nil {
+		lmcli.Fatal("Could not load messages: %v\n", err)
+	}
+
+	RenderConversation(ctx, append(existing, messages...), true)
+
+	var savedReplies []conversation.Message
+	if persist && len(messages) > 0 {
+		savedReplies, err = ctx.Conversations.Reply(to, messages...)
+		if err != nil {
+			lmcli.Warn("Could not save messages: %v\n", err)
 		}
 	}

-	allMessages := append(existing, toSend...)
-
-	RenderConversation(ctx, allMessages, true)
-
 	// render a message header with no contents
-	RenderMessage(ctx, (&model.Message{Role: model.MessageRoleAssistant}))
+	RenderMessage(ctx, (&conversation.Message{Role: api.MessageRoleAssistant}))

-	replyCallback := func(reply model.Message) {
+	var lastSavedMessage *conversation.Message
+	lastSavedMessage = to
+	if len(savedReplies) > 0 {
+		lastSavedMessage = &savedReplies[len(savedReplies)-1]
+	}
+
+	replyCallback := func(reply conversation.Message) {
 		if !persist {
 			return
 		}
-
-		reply.ConversationID = c.ID
-		err = ctx.Store.SaveMessage(&reply)
+		savedReplies, err = ctx.Conversations.Reply(lastSavedMessage, reply)
 		if err != nil {
 			lmcli.Warn("Could not save reply: %v\n", err)
 		}
+		lastSavedMessage = &savedReplies[0]
 	}

-	_, err = FetchAndShowCompletion(ctx, allMessages, replyCallback)
+	_, err = Prompt(ctx, append(existing, messages...), replyCallback)
 	if err != nil {
 		lmcli.Fatal("Error fetching LLM response: %v\n", err)
 	}
 }

-func FormatForExternalPrompt(messages []model.Message, system bool) string {
+func FormatForExternalPrompt(messages []conversation.Message, system bool) string {
 	sb := strings.Builder{}
 	for _, message := range messages {
 		if message.Content == "" {
 			continue
 		}
 		switch message.Role {
-		case model.MessageRoleAssistant, model.MessageRoleToolCall:
+		case api.MessageRoleAssistant, api.MessageRoleToolCall:
 			sb.WriteString("Assistant:\n\n")
-		case model.MessageRoleUser:
+		case api.MessageRoleUser:
 			sb.WriteString("User:\n\n")
 		default:
 			continue
@ -134,60 +165,76 @@ func FormatForExternalPrompt(messages []model.Message, system bool) string {
 	return sb.String()
 }

-func GenerateTitle(ctx *lmcli.Context, c *model.Conversation) (string, error) {
-	messages, err := ctx.Store.Messages(c)
-	if err != nil {
-		return "", err
-	}
-
-	const prompt = `Above is an excerpt from a conversation between a user and AI assistant. Please reply with a short title (no more than 8 words) that reflects the topic of the conversation, read from the user's perspective.
+func GenerateTitle(ctx *lmcli.Context, messages []conversation.Message) (string, error) {
+	const systemPrompt = `You will be shown a conversation between a user and an AI assistant. Your task is to generate a short title (8 words or less) for the provided conversation that reflects the conversation's topic. Your response is expected to be in JSON in the format shown below.

 Example conversation:

-"""
-User:
-
-  Hello!
-
-Assistant:
-
-  Hello! How may I assist you?
-"""
+[{"role": "user", "content": "Can you help me with my math homework?"},{"role": "assistant", "content": "Sure, what topic are you struggling with?"}]

 Example response:

-"""
-Title: A brief introduction
-"""
+{"title": "Help with math homework"}
 `
-	conversation := FormatForExternalPrompt(messages, false)
+	type msg struct {
+		Role    string
+		Content string
+	}

-	generateRequest := []model.Message{
+	var msgs []msg
+	for _, m := range messages {
+		switch m.Role {
+		case api.MessageRoleAssistant, api.MessageRoleUser:
+			msgs = append(msgs, msg{string(m.Role), m.Content})
+		}
+	}
+
+	// Serialize the conversation to JSON
+	jsonBytes, err := json.Marshal(msgs)
+	if err != nil {
+		return "", err
+	}
+
+	generateRequest := []conversation.Message{
 		{
-			Role:    model.MessageRoleUser,
-			Content: fmt.Sprintf("\"\"\"\n%s\n\"\"\"\n\n%s", conversation, prompt),
+			Role:    api.MessageRoleSystem,
+			Content: systemPrompt,
+		},
+		{
+			Role:    api.MessageRoleUser,
+			Content: string(jsonBytes),
 		},
 	}

-	completionProvider, err := ctx.GetCompletionProvider(*ctx.Config.Conversations.TitleGenerationModel)
+	m, _, p, err := ctx.GetModelProvider(
+		*ctx.Config.Conversations.TitleGenerationModel, "",
+	)
 	if err != nil {
 		return "", err
 	}

-	requestParams := model.RequestParameters{
-		Model:     *ctx.Config.Conversations.TitleGenerationModel,
+	requestParams := provider.RequestParameters{
+		Model:     m,
 		MaxTokens: 25,
 	}

-	response, err := completionProvider.CreateChatCompletion(context.Background(), requestParams, generateRequest, nil)
+	response, err := p.CreateChatCompletion(
+		context.Background(), requestParams, conversation.MessagesToAPI(generateRequest),
+	)
 	if err != nil {
 		return "", err
 	}

-	response = strings.TrimPrefix(response, "Title: ")
-	response = strings.Trim(response, "\"")
+	// Parse the JSON response
+	var jsonResponse struct {
+		Title string `json:"title"`
+	}
+	err = json.Unmarshal([]byte(response.Content), &jsonResponse)
+	if err != nil {
+		return "", err
+	}

-	return response, nil
+	return jsonResponse.Title, nil
 }

 // ShowWaitAnimation prints an animated ellipses to stdout until something is
@ -227,7 +274,7 @@ func ShowWaitAnimation(signal chan any) {
 // chunked) content is received on the channel, the waiting animation is
 // replaced by the content.
 // Blocks until the channel is closed.
-func ShowDelayedContent(content <-chan string) {
+func ShowDelayedContent(content <-chan provider.Chunk) {
 	waitSignal := make(chan any)
 	go ShowWaitAnimation(waitSignal)

@ -240,14 +287,14 @@ func ShowDelayedContent(content <-chan string) {
 			<-waitSignal
 			firstChunk = false
 		}
-		fmt.Print(chunk)
+		fmt.Print(chunk.Content)
 	}
 }

 // RenderConversation renders the given messages to TTY, with optional space
 // for a subsequent message. spaceForResponse controls how many '\n' characters
 // are printed immediately after the final message (1 if false, 2 if true)
-func RenderConversation(ctx *lmcli.Context, messages []model.Message, spaceForResponse bool) {
+func RenderConversation(ctx *lmcli.Context, messages []conversation.Message, spaceForResponse bool) {
 	l := len(messages)
 	for i, message := range messages {
 		RenderMessage(ctx, &message)
@ -258,7 +305,7 @@ func RenderConversation(ctx *lmcli.Context, messages []model.Message, spaceForRe
 	}
 }

-func RenderMessage(ctx *lmcli.Context, m *model.Message) {
+func RenderMessage(ctx *lmcli.Context, m *conversation.Message) {
 	var messageAge string
 	if m.CreatedAt.IsZero() {
 		messageAge = "now"
@ -270,11 +317,11 @@ func RenderMessage(ctx *lmcli.Context, m *model.Message) {
 	headerStyle := lipgloss.NewStyle().Bold(true)

 	switch m.Role {
-	case model.MessageRoleSystem:
+	case api.MessageRoleSystem:
 		headerStyle = headerStyle.Foreground(lipgloss.Color("9")) // bright red
-	case model.MessageRoleUser:
+	case api.MessageRoleUser:
 		headerStyle = headerStyle.Foreground(lipgloss.Color("10")) // bright green
-	case model.MessageRoleAssistant:
+	case api.MessageRoleAssistant:
 		headerStyle = headerStyle.Foreground(lipgloss.Color("12")) // bright blue
 	}

--- a/pkg/cmd/view.go
+++ b/pkg/cmd/view.go
@ -20,13 +20,13 @@ func ViewCmd(ctx *lmcli.Context) *cobra.Command {
 			}
 			return nil
 		},
-		RunE: func(cmd *cobra.Command, args []string)  error {
+		RunE: func(cmd *cobra.Command, args []string) error {
 			shortName := args[0]
 			conversation := cmdutil.LookupConversation(ctx, shortName)

-			messages, err := ctx.Store.Messages(conversation)
+			messages, err := ctx.Conversations.PathToLeaf(conversation.SelectedRoot)
 			if err != nil {
-				return fmt.Errorf("Could not retrieve messages for conversation: %s", conversation.Title)
+				return fmt.Errorf("Could not retrieve messages for conversation %s: %v", conversation.ShortName.String, err)
 			}

 			cmdutil.RenderConversation(ctx, messages, false)
@ -37,7 +37,7 @@ func ViewCmd(ctx *lmcli.Context) *cobra.Command {
 			if len(args) != 0 {
 				return nil, compMode
 			}
-			return ctx.Store.ConversationShortNameCompletions(toComplete), compMode
+			return ctx.Conversations.ConversationShortNameCompletions(toComplete), compMode
 		},
 	}

--- a/pkg/conversation/conversation.go
+++ b/pkg/conversation/conversation.go
@ -0,0 +1,99 @@
+package conversation
+
+import (
+	"database/sql"
+	"database/sql/driver"
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+)
+
+type Conversation struct {
+	ID             uint `gorm:"primaryKey"`
+	ShortName      sql.NullString
+	Title          string
+	SelectedRootID *uint
+	SelectedRoot   *Message  `gorm:"foreignKey:SelectedRootID"`
+	RootMessages   []Message `gorm:"-:all"`
+	LastMessageAt  time.Time
+}
+
+type MessageMeta struct {
+	GenerationProvider *string `json:"generation_provider,omitempty"`
+	GenerationModel    *string `json:"generation_model,omitempty"`
+}
+
+type Message struct {
+	ID        uint `gorm:"primaryKey"`
+	CreatedAt time.Time
+	Metadata  MessageMeta
+
+	ConversationID  *uint         `gorm:"index"`
+	Conversation    *Conversation `gorm:"foreignKey:ConversationID"`
+	ParentID        *uint
+	Parent          *Message  `gorm:"foreignKey:ParentID"`
+	Replies         []Message `gorm:"foreignKey:ParentID"`
+	SelectedReplyID *uint
+	SelectedReply   *Message `gorm:"foreignKey:SelectedReplyID"`
+
+	Role        api.MessageRole
+	Content     string
+	ToolCalls   ToolCalls   // a json array of tool calls (from the model)
+	ToolResults ToolResults // a json array of tool results
+}
+
+func (m *MessageMeta) Scan(value interface{}) error {
+	return json.Unmarshal(value.([]byte), m)
+}
+
+func (m MessageMeta) Value() (driver.Value, error) {
+	return json.Marshal(m)
+}
+
+type ToolCalls []api.ToolCall
+
+func (tc *ToolCalls) Scan(value any) (err error) {
+	s := value.(string)
+	if value == nil || s == "" {
+		*tc = nil
+		return
+	}
+	err = json.Unmarshal([]byte(s), tc)
+	return
+}
+
+func (tc ToolCalls) Value() (driver.Value, error) {
+	if len(tc) == 0 {
+		return "", nil
+	}
+	jsonBytes, err := json.Marshal(tc)
+	if err != nil {
+		return "", fmt.Errorf("Could not marshal ToolCalls to JSON: %v\n", err)
+	}
+	return string(jsonBytes), nil
+}
+
+type ToolResults []api.ToolResult
+
+func (tr *ToolResults) Scan(value any) (err error) {
+	s := value.(string)
+	if value == nil || s == "" {
+		*tr = nil
+		return
+	}
+	err = json.Unmarshal([]byte(s), tr)
+	return
+}
+
+func (tr ToolResults) Value() (driver.Value, error) {
+	if len(tr) == 0 {
+		return "", nil
+	}
+	jsonBytes, err := json.Marshal([]api.ToolResult(tr))
+	if err != nil {
+		return "", fmt.Errorf("Could not marshal ToolResults to JSON: %v\n", err)
+	}
+	return string(jsonBytes), nil
+}
--- a/pkg/conversation/repo.go
+++ b/pkg/conversation/repo.go
@ -0,0 +1,520 @@
+package conversation
+
+import (
+	"database/sql"
+	"errors"
+	"fmt"
+	"slices"
+	"strings"
+	"time"
+
+	sqids "github.com/sqids/sqids-go"
+	"gorm.io/gorm"
+)
+
+// Repo exposes low-level message and conversation management. See
+// Service for high-level helpers
+type Repo interface {
+	LoadConversationList() (ConversationList, error)
+
+	FindConversationByShortName(shortName string) (*Conversation, error)
+	ConversationShortNameCompletions(search string) []string
+	GetConversationByID(int uint) (*Conversation, error)
+	GetRootMessages(conversationID uint) ([]Message, error)
+
+	CreateConversation(title string) (*Conversation, error)
+	UpdateConversation(*Conversation) error
+	DeleteConversation(*Conversation) error
+	DeleteConversationById(id uint) error
+
+	GetMessageByID(messageID uint) (*Message, error)
+
+	SaveMessage(message Message) (*Message, error)
+	UpdateMessage(message *Message) error
+	DeleteMessage(message *Message, prune bool) error
+	CloneBranch(toClone Message) (*Message, uint, error)
+	Reply(to *Message, messages ...Message) ([]Message, error)
+
+	PathToRoot(message *Message) ([]Message, error)
+	PathToLeaf(message *Message) ([]Message, error)
+
+	// Retrieves and return the "selected thread" of the conversation.
+	// The "selected thread" of the conversation is a chain of messages
+	// starting from the Conversation's SelectedRoot Message, following each
+	// Message's SelectedReply until the tail Message is reached.
+	GetSelectedThread(*Conversation) ([]Message, error)
+
+	// Start a new conversation with the given messages
+	StartConversation(messages ...Message) (*Conversation, []Message, error)
+
+	CloneConversation(toClone Conversation) (*Conversation, uint, error)
+}
+
+type repo struct {
+	db    *gorm.DB
+	sqids *sqids.Sqids
+}
+
+func NewRepo(db *gorm.DB) (Repo, error) {
+	models := []any{
+		&Conversation{},
+		&Message{},
+	}
+
+	for _, x := range models {
+		err := db.AutoMigrate(x)
+		if err != nil {
+			return nil, fmt.Errorf("Could not perform database migrations: %v", err)
+		}
+	}
+
+	_sqids, _ := sqids.New(sqids.Options{MinLength: 4})
+	return &repo{db, _sqids}, nil
+}
+
+type ConversationListItem struct {
+	ID            uint
+	ShortName     string
+	Title         string
+	LastMessageAt time.Time
+}
+
+type ConversationList struct {
+	Total int
+	Items []ConversationListItem
+}
+
+// LoadConversationList loads existing conversations, ordered by the date
+// of their latest message, from most recent to oldest.
+func (s *repo) LoadConversationList() (ConversationList, error) {
+	list := ConversationList{}
+
+	var convos []Conversation
+	err := s.db.Order("last_message_at DESC").Find(&convos).Error
+	if err != nil {
+		return list, err
+	}
+
+	for _, c := range convos {
+		list.Items = append(list.Items, ConversationListItem{
+			ID:            c.ID,
+			ShortName:     c.ShortName.String,
+			Title:         c.Title,
+			LastMessageAt: c.LastMessageAt,
+		})
+	}
+
+	list.Total = len(list.Items)
+	return list, nil
+}
+
+func (s *repo) FindConversationByShortName(shortName string) (*Conversation, error) {
+	if shortName == "" {
+		return nil, errors.New("shortName is empty")
+	}
+	var conversation Conversation
+	err := s.db.Preload("SelectedRoot").Where("short_name = ?", shortName).Find(&conversation).Error
+	return &conversation, err
+}
+
+func (s *repo) ConversationShortNameCompletions(shortName string) []string {
+	var conversations []Conversation
+	// ignore error for completions
+	s.db.Find(&conversations)
+	completions := make([]string, 0, len(conversations))
+	for _, conversation := range conversations {
+		if shortName == "" || strings.HasPrefix(conversation.ShortName.String, shortName) {
+			completions = append(completions, fmt.Sprintf("%s\t%s", conversation.ShortName.String, conversation.Title))
+		}
+	}
+	return completions
+}
+
+func (s *repo) GetConversationByID(id uint) (*Conversation, error) {
+	var conversation Conversation
+	err := s.db.Preload("SelectedRoot").Where("id = ?", id).Find(&conversation).Error
+	if err != nil {
+		return nil, fmt.Errorf("Cannot get conversation %d: %v", id, err)
+	}
+
+	rootMessages, err := s.GetRootMessages(id)
+	if err != nil {
+		return nil, fmt.Errorf("Could not load conversation's root messages %d: %v", id, err)
+	}
+	conversation.RootMessages = rootMessages
+	return &conversation, nil
+}
+
+func (s *repo) CreateConversation(title string) (*Conversation, error) {
+	// Create the new conversation
+	c := &Conversation{Title: title}
+	err := s.db.Create(c).Error
+	if err != nil {
+		return nil, err
+	}
+	// Generate and save its "short name"
+	shortName, _ := s.sqids.Encode([]uint64{uint64(c.ID)})
+	c.ShortName = sql.NullString{String: shortName, Valid: true}
+	err = s.db.Updates(c).Error
+	if err != nil {
+		return nil, err
+	}
+	return c, nil
+}
+
+func (s *repo) UpdateConversation(c *Conversation) error {
+	if c == nil || c.ID == 0 {
+		return fmt.Errorf("Conversation is nil or invalid (missing ID)")
+	}
+	return s.db.Updates(c).Error
+}
+
+func (s *repo) DeleteConversation(c *Conversation) error {
+	if c == nil || c.ID == 0 {
+		return fmt.Errorf("Conversation is nil or invalid (missing ID)")
+	}
+	return s.DeleteConversationById(c.ID)
+}
+
+func (s *repo) DeleteConversationById(id uint) error {
+	if id == 0 {
+		return fmt.Errorf("Invalid conversation ID: %d", id)
+	}
+	err := s.db.Where("conversation_id = ?", id).Delete(&Message{}).Error
+	if err != nil {
+		return err
+	}
+	return s.db.Where("id = ?", id).Delete(&Conversation{}).Error
+}
+
+func (s *repo) SaveMessage(m Message) (*Message, error) {
+	if m.Conversation == nil {
+		return nil, fmt.Errorf("Can't save a message without a conversation (this is a bug)")
+	}
+	newMessage := m
+	newMessage.ID = 0
+	newMessage.CreatedAt = time.Now()
+	return &newMessage, s.db.Create(&newMessage).Error
+}
+
+func (s *repo) UpdateMessage(m *Message) error {
+	if m == nil || m.ID == 0 {
+		return fmt.Errorf("Message is nil or invalid (missing ID)")
+	}
+	return s.db.Updates(m).Error
+}
+
+func (s *repo) DeleteMessage(message *Message, prune bool) error {
+	return s.db.Delete(&message).Error
+}
+
+func (s *repo) GetMessageByID(messageID uint) (*Message, error) {
+	var message Message
+	err := s.db.Preload("Parent").Preload("Replies").Preload("SelectedReply").Where("id = ?", messageID).Find(&message).Error
+	return &message, err
+}
+
+// Reply to a message with a series of messages (each followed by the next)
+func (s *repo) Reply(to *Message, messages ...Message) ([]Message, error) {
+	var savedMessages []Message
+
+	err := s.db.Transaction(func(tx *gorm.DB) error {
+		currentParent := to
+		for i := range messages {
+			parent := currentParent
+			message := messages[i]
+			message.Parent = parent
+			message.Conversation = parent.Conversation
+			message.ID = 0
+			message.CreatedAt = time.Time{}
+
+			if err := tx.Create(&message).Error; err != nil {
+				return err
+			}
+
+			// update parent selected reply
+			parent.Replies = append(parent.Replies, message)
+			parent.SelectedReply = &message
+			if err := tx.Model(parent).Update("selected_reply_id", message.ID).Error; err != nil {
+				return err
+			}
+
+			savedMessages = append(savedMessages, message)
+			currentParent = &message
+		}
+		return nil
+	})
+
+	if err != nil {
+		return savedMessages, err
+	}
+
+	to.Conversation.LastMessageAt = savedMessages[len(savedMessages)-1].CreatedAt
+	err = s.UpdateConversation(to.Conversation)
+	return savedMessages, err
+}
+
+// CloneBranch returns a deep clone of the given message and its replies, returning
+// a new message object. The new message will be attached to the same parent as
+// the messageToClone
+func (s *repo) CloneBranch(messageToClone Message) (*Message, uint, error) {
+	newMessage := messageToClone
+	newMessage.ID = 0
+	newMessage.Replies = nil
+	newMessage.SelectedReplyID = nil
+	newMessage.SelectedReply = nil
+
+	originalReplies := messageToClone.Replies
+
+	if err := s.db.Create(&newMessage).Error; err != nil {
+		return nil, 0, fmt.Errorf("Could not clone message: %s", err)
+	}
+
+	var replyCount uint = 0
+	for _, reply := range originalReplies {
+		replyCount++
+
+		newReply := reply
+		newReply.ConversationID = messageToClone.ConversationID
+		newReply.ParentID = &newMessage.ID
+		newReply.Parent = &newMessage
+
+		res, c, err := s.CloneBranch(newReply)
+		if err != nil {
+			return nil, 0, err
+		}
+		newMessage.Replies = append(newMessage.Replies, *res)
+		replyCount += c
+
+		if reply.ID == *messageToClone.SelectedReplyID {
+			newMessage.SelectedReplyID = &res.ID
+			if err := s.UpdateMessage(&newMessage); err != nil {
+				return nil, 0, fmt.Errorf("Could not update parent select reply ID: %v", err)
+			}
+		}
+	}
+	return &newMessage, replyCount, nil
+}
+
+func fetchMessages(db *gorm.DB) ([]Message, error) {
+	var messages []Message
+	if err := db.Preload("Conversation").Find(&messages).Error; err != nil {
+		return nil, fmt.Errorf("Could not fetch messages: %v", err)
+	}
+
+	messageMap := make(map[uint]Message)
+	for i, message := range messages {
+		messageMap[messages[i].ID] = message
+	}
+
+	// Create a map to store replies by their parent ID
+	repliesMap := make(map[uint][]Message)
+	for i, message := range messages {
+		if messages[i].ParentID != nil {
+			repliesMap[*messages[i].ParentID] = append(repliesMap[*messages[i].ParentID], message)
+		}
+	}
+
+	// Assign replies, parent, and selected reply to each message
+	for i := range messages {
+		if replies, exists := repliesMap[messages[i].ID]; exists {
+			messages[i].Replies = make([]Message, len(replies))
+			for j, m := range replies {
+				messages[i].Replies[j] = m
+			}
+		}
+		if messages[i].ParentID != nil {
+			if parent, exists := messageMap[*messages[i].ParentID]; exists {
+				messages[i].Parent = &parent
+			}
+		}
+		if messages[i].SelectedReplyID != nil {
+			if selectedReply, exists := messageMap[*messages[i].SelectedReplyID]; exists {
+				messages[i].SelectedReply = &selectedReply
+			}
+		}
+	}
+	return messages, nil
+}
+
+func (r repo) GetRootMessages(conversationID uint) ([]Message, error) {
+	var rootMessages []Message
+	err := r.db.Where("conversation_id = ? AND parent_id IS NULL", conversationID).Find(&rootMessages).Error
+	if err != nil {
+		return nil, fmt.Errorf("Could not retrieve root messages for conversation %d: %v", conversationID, err)
+	}
+	return rootMessages, nil
+}
+
+func (s *repo) buildPath(message *Message, getNext func(*Message) *uint) ([]Message, error) {
+	var messages []Message
+	messages, err := fetchMessages(s.db.Where("conversation_id = ?", message.ConversationID))
+	if err != nil {
+		return nil, err
+	}
+
+	// Create a map to store messages by their ID
+	messageMap := make(map[uint]*Message, len(messages))
+	for i := range messages {
+		messageMap[messages[i].ID] = &messages[i]
+	}
+
+	// Construct Replies
+	repliesMap := make(map[uint][]*Message, len(messages))
+	for _, m := range messageMap {
+		if m.ParentID == nil {
+			continue
+		}
+		if p, ok := messageMap[*m.ParentID]; ok {
+			repliesMap[p.ID] = append(repliesMap[p.ID], m)
+		}
+	}
+
+	// Add replies to messages
+	for _, m := range messageMap {
+		if replies, ok := repliesMap[m.ID]; ok {
+			m.Replies = make([]Message, len(replies))
+			for idx, reply := range replies {
+				m.Replies[idx] = *reply
+			}
+		}
+	}
+
+	// Build the path
+	var path []Message
+	nextID := &message.ID
+
+	for {
+		current, exists := messageMap[*nextID]
+		if !exists {
+			return nil, fmt.Errorf("Message with ID %d not found in conversation", *nextID)
+		}
+
+		path = append(path, *current)
+
+		nextID = getNext(current)
+		if nextID == nil {
+			break
+		}
+	}
+
+	return path, nil
+}
+
+// PathToRoot traverses the provided message's Parent until reaching the tree
+// root and returns a slice of all messages traversed in chronological order
+// (starting with the root and ending with the message provided)
+func (s *repo) PathToRoot(message *Message) ([]Message, error) {
+	if message == nil || message.ID <= 0 {
+		return nil, fmt.Errorf("Message is nil or has invalid ID")
+	}
+
+	path, err := s.buildPath(message, func(m *Message) *uint {
+		return m.ParentID
+	})
+	if err != nil {
+		return nil, err
+	}
+	slices.Reverse(path)
+
+	return path, nil
+}
+
+// PathToLeaf traverses the provided message's SelectedReply until reaching a
+// tree leaf and returns a slice of all messages traversed in chronological
+// order (starting with the message provided and ending with the leaf)
+func (s *repo) PathToLeaf(message *Message) ([]Message, error) {
+	if message == nil || message.ID <= 0 {
+		return nil, fmt.Errorf("Message is nil or has invalid ID")
+	}
+
+	return s.buildPath(message, func(m *Message) *uint {
+		return m.SelectedReplyID
+	})
+}
+
+func (s *repo) StartConversation(messages ...Message) (*Conversation, []Message, error) {
+	if len(messages) == 0 {
+		return nil, nil, fmt.Errorf("Must provide at least 1 message")
+	}
+
+	// Create new conversation
+	conversation, err := s.CreateConversation("")
+	if err != nil {
+		return nil, nil, err
+	}
+	messages[0].Conversation = conversation
+
+	// Create first message
+	firstMessage, err := s.SaveMessage(messages[0])
+	if err != nil {
+		return nil, nil, err
+	}
+	messages[0] = *firstMessage
+
+	// Update conversation's selected root message
+	conversation.RootMessages = []Message{messages[0]}
+	conversation.SelectedRoot = &messages[0]
+	conversation.LastMessageAt = messages[0].CreatedAt
+
+	// Add additional replies to conversation
+	if len(messages) > 1 {
+		newMessages, err := s.Reply(&messages[0], messages[1:]...)
+		if err != nil {
+			return nil, nil, err
+		}
+		messages = append([]Message{messages[0]}, newMessages...)
+		conversation.LastMessageAt = messages[len(messages)-1].CreatedAt
+	}
+
+	err = s.UpdateConversation(conversation)
+	return conversation, messages, err
+}
+
+// CloneConversation clones the given conversation and all of its meesages
+func (s *repo) CloneConversation(toClone Conversation) (*Conversation, uint, error) {
+	rootMessages, err := s.GetRootMessages(toClone.ID)
+	if err != nil {
+		return nil, 0, fmt.Errorf("Could not create clone: %v", err)
+	}
+
+	clone, err := s.CreateConversation(toClone.Title + " - Clone")
+	if err != nil {
+		return nil, 0, fmt.Errorf("Could not create clone: %v", err)
+	}
+
+	var errors []error
+	var messageCnt uint = 0
+	for _, root := range rootMessages {
+		messageCnt++
+		newRoot := root
+		newRoot.ConversationID = &clone.ID
+
+		cloned, count, err := s.CloneBranch(newRoot)
+		if err != nil {
+			errors = append(errors, err)
+			continue
+		}
+		messageCnt += count
+
+		if root.ID == *toClone.SelectedRootID {
+			clone.SelectedRootID = &cloned.ID
+			if err := s.UpdateConversation(clone); err != nil {
+				errors = append(errors, fmt.Errorf("Could not set selected root on clone: %v", err))
+			}
+		}
+	}
+
+	if len(errors) > 0 {
+		return nil, 0, fmt.Errorf("Messages failed to be cloned: %v", errors)
+	}
+
+	return clone, messageCnt, nil
+}
+
+func (s *repo) GetSelectedThread(c *Conversation) ([]Message, error) {
+	if c.SelectedRoot == nil {
+		return nil, fmt.Errorf("No SelectedRoot on conversation - this is a bug")
+	}
+	return s.PathToLeaf(c.SelectedRoot)
+}
--- a/pkg/conversation/tools.go
+++ b/pkg/conversation/tools.go
@ -0,0 +1,55 @@
+package conversation
+
+import (
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+)
+
+// ApplySystemPrompt updates the contents of an existing system Message if it
+// exists, or returns a new slice with the system Message prepended.
+func ApplySystemPrompt(m []Message, system string, force bool) []Message {
+	if len(m) > 0 && m[0].Role == api.MessageRoleSystem {
+		if force {
+			m[0].Content = system
+		}
+		return m
+	} else {
+		return append([]Message{{
+			Role:    api.MessageRoleSystem,
+			Content: system,
+		}}, m...)
+	}
+}
+
+func MessageToAPI(m Message) api.Message {
+	return api.Message{
+		Role:        m.Role,
+		Content:     m.Content,
+		ToolCalls:   m.ToolCalls,
+		ToolResults: m.ToolResults,
+	}
+}
+
+func MessagesToAPI(messages []Message) []api.Message {
+	ret := make([]api.Message, 0, len(messages))
+	for _, m := range messages {
+		ret = append(ret, MessageToAPI(m))
+	}
+	return ret
+}
+
+func MessageFromAPI(m api.Message) Message {
+	return Message{
+		Role:        m.Role,
+		Content:     m.Content,
+		ToolCalls:   m.ToolCalls,
+		ToolResults: m.ToolResults,
+	}
+}
+
+func MessagesFromAPI(messages []api.Message) []Message {
+	ret := make([]Message, 0, len(messages))
+	for _, m := range messages {
+		ret = append(ret, MessageFromAPI(m))
+	}
+	return ret
+}
--- a/pkg/lmcli/config.go
+++ b/pkg/lmcli/config.go
@ -5,34 +5,39 @@ import (
 	"os"

 	"git.mlow.ca/mlow/lmcli/pkg/util"
-	"github.com/go-yaml/yaml"
+	"gopkg.in/yaml.v3"
 )

 type Config struct {
 	Defaults *struct {
-		SystemPrompt *string  `yaml:"systemPrompt" default:"You are a helpful assistant."`
-		MaxTokens    *int     `yaml:"maxTokens" default:"256"`
-		Temperature  *float32 `yaml:"temperature" default:"0.7"`
-		Model        *string  `yaml:"model" default:"gpt-4"`
+		Model            *string  `yaml:"model" default:"gpt-4"`
+		MaxTokens        *int     `yaml:"maxTokens" default:"256"`
+		Temperature      *float32 `yaml:"temperature" default:"0.2"`
+		SystemPrompt     string   `yaml:"systemPrompt,omitempty"`
+		SystemPromptFile string   `yaml:"systemPromptFile,omitempty"`
+		Agent            string   `yaml:"agent"`
 	} `yaml:"defaults"`
 	Conversations *struct {
 		TitleGenerationModel *string `yaml:"titleGenerationModel" default:"gpt-3.5-turbo"`
 	} `yaml:"conversations"`
-	Tools *struct {
-		EnabledTools *[]string `yaml:"enabledTools"`
-	} `yaml:"tools"`
-	OpenAI *struct {
-		APIKey *string   `yaml:"apiKey" default:"your_key_here"`
-		Models *[]string `yaml:"models"`
-	} `yaml:"openai"`
-	Anthropic *struct {
-		APIKey *string   `yaml:"apiKey" default:"your_key_here"`
-		Models *[]string `yaml:"models"`
-	} `yaml:"anthropic"`
 	Chroma *struct {
 		Style     *string `yaml:"style" default:"onedark"`
 		Formatter *string `yaml:"formatter" default:"terminal16m"`
 	} `yaml:"chroma"`
+	Agents []*struct {
+		Name         string   `yaml:"name"`
+		SystemPrompt string   `yaml:"systemPrompt"`
+		Tools        []string `yaml:"tools"`
+	} `yaml:"agents"`
+	Providers []*struct {
+		Name    string            `yaml:"name,omitempty"`
+		Display string            `yaml:"display,omitempty"`
+		Kind    string            `yaml:"kind"`
+		BaseURL string            `yaml:"baseUrl,omitempty"`
+		APIKey  string            `yaml:"apiKey,omitempty"`
+		Models  []string          `yaml:"models"`
+		Headers map[string]string `yaml:"headers"`
+	} `yaml:"providers"`
 }

 func NewConfig(configFile string) (*Config, error) {
@ -60,8 +65,9 @@ func NewConfig(configFile string) (*Config, error) {
 		if err != nil {
 			return nil, fmt.Errorf("Could not open config file for writing: %v", err)
 		}
-		bytes, _ := yaml.Marshal(c)
-		_, err = file.Write(bytes)
+		encoder := yaml.NewEncoder(file)
+		encoder.SetIndent(2)
+		err = encoder.Encode(c)
 		if err != nil {
 			return nil, fmt.Errorf("Could not save default configuration: %v", err)
 		}
--- a/pkg/lmcli/lmcli.go
+++ b/pkg/lmcli/lmcli.go
@ -1,86 +1,231 @@
 package lmcli

 import (
+	"errors"
 	"fmt"
+	"io/fs"
+	"log"
 	"os"
 	"path/filepath"
+	"strings"
+	"time"

-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/provider"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/provider/anthropic"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/provider/openai"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/tools"
+	"git.mlow.ca/mlow/lmcli/pkg/agents"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+	"git.mlow.ca/mlow/lmcli/pkg/provider/anthropic"
+	"git.mlow.ca/mlow/lmcli/pkg/provider/google"
+	"git.mlow.ca/mlow/lmcli/pkg/provider/ollama"
+	"git.mlow.ca/mlow/lmcli/pkg/provider/openai"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/util"
 	"git.mlow.ca/mlow/lmcli/pkg/util/tty"
 	"gorm.io/driver/sqlite"
 	"gorm.io/gorm"
+	"gorm.io/gorm/logger"
 )

-type Context struct {
-	Config *Config
-	Store  ConversationStore
+type Agent struct {
+	Name         string
+	SystemPrompt string
+	Toolbox      []api.ToolSpec
+}

+type Context struct {
+	// high level app configuration, may be mutated at runtime
+	Config Config
+	Conversations conversation.Repo
 	Chroma *tty.ChromaHighlighter
-	EnabledTools []model.Tool
 }

 func NewContext() (*Context, error) {
 	configFile := filepath.Join(configDir(), "config.yaml")
 	config, err := NewConfig(configFile)
 	if err != nil {
-		Fatal("%v\n", err)
+		return nil, err
 	}

-	databaseFile := filepath.Join(dataDir(), "conversations.db")
-	db, err := gorm.Open(sqlite.Open(databaseFile), &gorm.Config{})
+	store, err := getConversationService()
 	if err != nil {
-		return nil, fmt.Errorf("Error establishing connection to store: %v", err)
-	}
-	store, err := NewSQLStore(db)
-	if err != nil {
-		Fatal("%v\n", err)
+		return nil, err
 	}

 	chroma := tty.NewChromaHighlighter("markdown", *config.Chroma.Formatter, *config.Chroma.Style)
+	return &Context{*config, store, chroma}, nil
+}

-	var enabledTools []model.Tool
-	for _, toolName := range *config.Tools.EnabledTools {
-		tool, ok := tools.AvailableTools[toolName]
-		if ok {
-			enabledTools = append(enabledTools, tool)
+func createOrOpenAppend(path string) (*os.File, error) {
+	var file *os.File
+	if _, err := os.Stat(path); errors.Is(err, os.ErrNotExist) {
+		file, err = os.Create(path)
+		if err != nil {
+			return nil, err
+		}
+	} else {
+		file, err = os.OpenFile(path, os.O_APPEND, fs.ModeAppend)
+		if err != nil {
+			return nil, err
 		}
 	}
+	return file, nil
+}

-	return &Context{config, store, chroma, enabledTools}, nil
+func getConversationService() (conversation.Repo, error) {
+	databaseFile := filepath.Join(dataDir(), "conversations.db")
+	gormLogFile, err := createOrOpenAppend(filepath.Join(dataDir(), "database.log"))
+	if err != nil {
+		return nil, fmt.Errorf("Could not open database log file: %v", err)
+	}
+	db, err := gorm.Open(sqlite.Open(databaseFile), &gorm.Config{
+		Logger: logger.New(log.New(gormLogFile, "\n", log.LstdFlags), logger.Config{
+			SlowThreshold:             200 * time.Millisecond,
+			LogLevel:                  logger.Info,
+			IgnoreRecordNotFoundError: false,
+			Colorful:                  true,
+		}),
+	})
+	if err != nil {
+		return nil, fmt.Errorf("Error establishing connection to store: %v", err)
+	}
+	repo, err := conversation.NewRepo(db)
+	if err != nil {
+		return nil, err
+	}
+	return repo, nil
 }

 func (c *Context) GetModels() (models []string) {
-	for _, m := range *c.Config.Anthropic.Models {
-		models = append(models, m)
+	modelCounts := make(map[string]int)
+	for _, p := range c.Config.Providers {
+		name := p.Kind
+		if p.Name != "" {
+			name = p.Name
+		}
+
+		for _, m := range p.Models {
+			modelCounts[m]++
+			models = append(models, fmt.Sprintf("%s@%s", m, name))
+		}
 	}
-	for _, m := range *c.Config.OpenAI.Models {
-		models = append(models, m)
+
+	for m, c := range modelCounts {
+		if c == 1 {
+			models = append(models, m)
+		}
+	}
+
+	return
+}
+
+func (c *Context) GetAgents() (agents []string) {
+	for _, p := range c.Config.Agents {
+		agents = append(agents, p.Name)
 	}
 	return
 }

-func (c *Context) GetCompletionProvider(model string) (provider.ChatCompletionClient, error) {
-	for _, m := range *c.Config.Anthropic.Models {
-		if m == model {
-			anthropic := &anthropic.AnthropicClient{
-				APIKey: *c.Config.Anthropic.APIKey,
+func (c *Context) GetAgent(name string) *Agent {
+	if name == "" || name == "none" {
+		return nil
+	}
+
+	for _, a := range c.Config.Agents {
+		if name != a.Name {
+			continue
+		}
+
+		var enabledTools []api.ToolSpec
+		for _, toolName := range a.Tools {
+			tool, ok := agents.AvailableTools[toolName]
+			if ok {
+				enabledTools = append(enabledTools, tool)
 			}
-			return anthropic, nil
+		}
+
+		return &Agent{
+			Name:         a.Name,
+			SystemPrompt: a.SystemPrompt,
+			Toolbox:      enabledTools,
 		}
 	}
-	for _, m := range *c.Config.OpenAI.Models {
-		if m == model {
-			openai := &openai.OpenAIClient{
-				APIKey: *c.Config.OpenAI.APIKey,
+	return nil
+}
+
+func (c *Context) DefaultSystemPrompt() string {
+	if c.Config.Defaults.SystemPromptFile != "" {
+		content, err := util.ReadFileContents(c.Config.Defaults.SystemPromptFile)
+		if err != nil {
+			Fatal("Could not read file contents at %s: %v\n", c.Config.Defaults.SystemPromptFile, err)
+		}
+		return content
+	}
+	return c.Config.Defaults.SystemPrompt
+}
+
+func (c *Context) GetModelProvider(model string, provider string) (string, string, provider.ChatCompletionProvider, error) {
+	parts := strings.Split(model, "@")
+
+	if provider == "" && len(parts) > 1 {
+		model = parts[0]
+		provider = parts[1]
+	}
+
+	for _, p := range c.Config.Providers {
+		name := p.Kind
+		if p.Name != "" {
+			name = p.Name
+		}
+
+		if provider != "" && name != provider {
+			continue
+		}
+
+		for _, m := range p.Models {
+			if m == model {
+				switch p.Kind {
+				case "anthropic":
+					url := "https://api.anthropic.com"
+					if p.BaseURL != "" {
+						url = p.BaseURL
+					}
+					return model, name, &anthropic.AnthropicClient{
+						BaseURL: url,
+						APIKey:  p.APIKey,
+					}, nil
+				case "google":
+					url := "https://generativelanguage.googleapis.com"
+					if p.BaseURL != "" {
+						url = p.BaseURL
+					}
+					return model, name, &google.Client{
+						BaseURL: url,
+						APIKey:  p.APIKey,
+					}, nil
+				case "ollama":
+					url := "http://localhost:11434/api"
+					if p.BaseURL != "" {
+						url = p.BaseURL
+					}
+					return model, name, &ollama.OllamaClient{
+						BaseURL: url,
+					}, nil
+				case "openai":
+					url := "https://api.openai.com"
+					if p.BaseURL != "" {
+						url = p.BaseURL
+					}
+					return model, name, &openai.OpenAIClient{
+						BaseURL: url,
+						APIKey:  p.APIKey,
+						Headers: p.Headers,
+					}, nil
+				default:
+					return "", "", nil, fmt.Errorf("unknown provider kind: %s", p.Kind)
+				}
 			}
-			return openai, nil
 		}
 	}
-	return nil, fmt.Errorf("unknown model: %s", model)
+	return "", "", nil, fmt.Errorf("unknown model: %s", model)
 }

 func configDir() string {
--- a/pkg/lmcli/model/conversation.go
+++ b/pkg/lmcli/model/conversation.go
@ -1,58 +0,0 @@
-package model
-
-import (
-	"database/sql"
-	"time"
-)
-
-type MessageRole string
-
-const (
-	MessageRoleSystem     MessageRole = "system"
-	MessageRoleUser       MessageRole = "user"
-	MessageRoleAssistant  MessageRole = "assistant"
-	MessageRoleToolCall   MessageRole = "tool_call"
-	MessageRoleToolResult MessageRole = "tool_result"
-)
-
-type Message struct {
-	ID             uint `gorm:"primaryKey"`
-	ConversationID uint `gorm:"foreignKey:ConversationID"`
-	Content        string
-	Role           MessageRole
-	CreatedAt      time.Time
-	ToolCalls      ToolCalls   // a json array of tool calls (from the modl)
-	ToolResults    ToolResults // a json array of tool results
-}
-
-type Conversation struct {
-	ID        uint `gorm:"primaryKey"`
-	ShortName sql.NullString
-	Title     string
-}
-
-type RequestParameters struct {
-	Model       string
-	MaxTokens   int
-	Temperature float32
-	TopP        float32
-
-	SystemPrompt string
-	ToolBag      []Tool
-}
-
-// FriendlyRole returns a human friendly signifier for the message's role.
-func (m *MessageRole) FriendlyRole() string {
-	var friendlyRole string
-	switch *m {
-	case MessageRoleUser:
-		friendlyRole = "You"
-	case MessageRoleSystem:
-		friendlyRole = "System"
-	case MessageRoleAssistant:
-		friendlyRole = "Assistant"
-	default:
-		friendlyRole = string(*m)
-	}
-	return friendlyRole
-}
--- a/pkg/lmcli/model/tool.go
+++ b/pkg/lmcli/model/tool.go
@ -1,98 +0,0 @@
-package model
-
-import (
-	"database/sql/driver"
-	"encoding/json"
-	"fmt"
-)
-
-type Tool struct {
-	Name        string
-	Description string
-	Parameters  []ToolParameter
-	Impl        func(*Tool, map[string]interface{}) (string, error)
-}
-
-type ToolParameter struct {
-	Name        string   `json:"name"`
-	Type        string   `json:"type"` // "string", "integer", "boolean"
-	Required    bool     `json:"required"`
-	Description string   `json:"description"`
-	Enum        []string `json:"enum,omitempty"`
-}
-
-type ToolCall struct {
-	ID         string                 `json:"id"`
-	Name       string                 `json:"name"`
-	Parameters map[string]interface{} `json:"parameters"`
-}
-
-type ToolCalls []ToolCall
-
-func (tc *ToolCalls) Scan(value any) (err error) {
-	s := value.(string)
-	if value == nil || s == "" {
-		*tc = nil
-		return
-	}
-	err = json.Unmarshal([]byte(s), tc)
-	return
-}
-
-func (tc ToolCalls) Value() (driver.Value, error) {
-	if len(tc) == 0 {
-		return "", nil
-	}
-	jsonBytes, err := json.Marshal(tc)
-	if err != nil {
-		return "", fmt.Errorf("Could not marshal ToolCalls to JSON: %v\n", err)
-	}
-	return string(jsonBytes), nil
-}
-
-type ToolResult struct {
-	ToolCallID string `json:"toolCallID"`
-	ToolName   string `json:"toolName,omitempty"`
-	Result     string `json:"result,omitempty"`
-}
-
-type ToolResults []ToolResult
-
-func (tr *ToolResults) Scan(value any) (err error) {
-	s := value.(string)
-	if value == nil || s == "" {
-		*tr = nil
-		return
-	}
-	err = json.Unmarshal([]byte(s), tr)
-	return
-}
-
-func (tr ToolResults) Value() (driver.Value, error) {
-	if len(tr) == 0 {
-		return "", nil
-	}
-	jsonBytes, err := json.Marshal([]ToolResult(tr))
-	if err != nil {
-		return "", fmt.Errorf("Could not marshal ToolResults to JSON: %v\n", err)
-	}
-	return string(jsonBytes), nil
-}
-
-type CallResult struct {
-	Message string `json:"message"`
-	Result  any    `json:"result,omitempty"`
-}
-
-func (r CallResult) ToJson() (string, error) {
-	if r.Message == "" {
-		// When message not supplied, assume success
-		r.Message = "success"
-	}
-
-	jsonBytes, err := json.Marshal(r)
-	if err != nil {
-		return "", fmt.Errorf("Could not marshal CallResult to JSON: %v\n", err)
-	}
-	return string(jsonBytes), nil
-}
--- a/pkg/lmcli/provider/anthropic/anthropic.go
+++ b/pkg/lmcli/provider/anthropic/anthropic.go
@ -1,348 +0,0 @@
-package anthropic
-
-import (
-	"bufio"
-	"bytes"
-	"context"
-	"encoding/json"
-	"encoding/xml"
-	"fmt"
-	"net/http"
-	"strings"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/provider"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/tools"
-)
-
-type AnthropicClient struct {
-	APIKey string
-}
-
-type Message struct {
-	Role    string `json:"role"`
-	Content string `json:"content"`
-}
-
-type Request struct {
-	Model         string    `json:"model"`
-	Messages      []Message `json:"messages"`
-	System        string    `json:"system,omitempty"`
-	MaxTokens     int       `json:"max_tokens,omitempty"`
-	StopSequences []string  `json:"stop_sequences,omitempty"`
-	Stream        bool      `json:"stream,omitempty"`
-	Temperature   float32   `json:"temperature,omitempty"`
-	//TopP          float32   `json:"top_p,omitempty"`
-	//TopK          float32   `json:"top_k,omitempty"`
-}
-
-type OriginalContent struct {
-	Type string `json:"type"`
-	Text string `json:"text"`
-}
-
-type Response struct {
-	Id      string            `json:"id"`
-	Type    string            `json:"type"`
-	Role    string            `json:"role"`
-	Content []OriginalContent `json:"content"`
-}
-
-const FUNCTION_STOP_SEQUENCE = "</function_calls>"
-
-func buildRequest(params model.RequestParameters, messages []model.Message) Request {
-	requestBody := Request{
-		Model:       params.Model,
-		Messages:    make([]Message, len(messages)),
-		System:      params.SystemPrompt,
-		MaxTokens:   params.MaxTokens,
-		Temperature: params.Temperature,
-		Stream:      false,
-
-		StopSequences: []string{
-			FUNCTION_STOP_SEQUENCE,
-			"\n\nHuman:",
-		},
-	}
-
-	startIdx := 0
-	if len(messages) > 0 && messages[0].Role == model.MessageRoleSystem {
-		requestBody.System = messages[0].Content
-		requestBody.Messages = requestBody.Messages[1:]
-		startIdx = 1
-	}
-
-	if len(params.ToolBag) > 0 {
-		if len(requestBody.System) > 0 {
-			// add a divider between existing system prompt and tools
-			requestBody.System += "\n\n---\n\n"
-		}
-		requestBody.System += buildToolsSystemPrompt(params.ToolBag)
-	}
-
-	for i, msg := range messages[startIdx:] {
-		message := &requestBody.Messages[i]
-
-		switch msg.Role {
-		case model.MessageRoleToolCall:
-			message.Role = "assistant"
-			if msg.Content != "" {
-				message.Content = msg.Content
-			}
-			xmlFuncCalls := convertToolCallsToXMLFunctionCalls(msg.ToolCalls)
-			xmlString, err := xmlFuncCalls.XMLString()
-			if err != nil {
-				panic("Could not serialize []ToolCall to XMLFunctionCall")
-			}
-			if len(message.Content) > 0 {
-				message.Content += fmt.Sprintf("\n\n%s", xmlString)
-			} else {
-				message.Content = xmlString
-			}
-		case model.MessageRoleToolResult:
-			xmlFuncResults := convertToolResultsToXMLFunctionResult(msg.ToolResults)
-			xmlString, err := xmlFuncResults.XMLString()
-			if err != nil {
-				panic("Could not serialize []ToolResult to XMLFunctionResults")
-			}
-			message.Role = "user"
-			message.Content = xmlString
-		default:
-			message.Role = string(msg.Role)
-			message.Content = msg.Content
-		}
-	}
-	return requestBody
-}
-
-func sendRequest(ctx context.Context, c *AnthropicClient, r Request) (*http.Response, error) {
-	url := "https://api.anthropic.com/v1/messages"
-
-	jsonBody, err := json.Marshal(r)
-	if err != nil {
-		return nil, fmt.Errorf("failed to marshal request body: %v", err)
-	}
-
-	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonBody))
-	if err != nil {
-		return nil, fmt.Errorf("failed to create HTTP request: %v", err)
-	}
-
-	req.Header.Set("x-api-key", c.APIKey)
-	req.Header.Set("anthropic-version", "2023-06-01")
-	req.Header.Set("content-type", "application/json")
-
-	client := &http.Client{}
-	resp, err := client.Do(req)
-	if err != nil {
-		return nil, fmt.Errorf("failed to send HTTP request: %v", err)
-	}
-
-	return resp, nil
-}
-
-func (c *AnthropicClient) CreateChatCompletion(
-	ctx context.Context,
-	params model.RequestParameters,
-	messages []model.Message,
-	callback provider.ReplyCallback,
-) (string, error) {
-	request := buildRequest(params, messages)
-
-	resp, err := sendRequest(ctx, c, request)
-	if err != nil {
-		return "", err
-	}
-	defer resp.Body.Close()
-
-	var response Response
-	err = json.NewDecoder(resp.Body).Decode(&response)
-	if err != nil {
-		return "", fmt.Errorf("failed to decode response: %v", err)
-	}
-
-	sb := strings.Builder{}
-	for _, content := range response.Content {
-		var reply model.Message
-		switch content.Type {
-		case "text":
-			reply = model.Message{
-				Role:    model.MessageRoleAssistant,
-				Content: content.Text,
-			}
-			sb.WriteString(reply.Content)
-		default:
-			return "", fmt.Errorf("unsupported message type: %s", content.Type)
-		}
-		if callback != nil {
-			callback(reply)
-		}
-	}
-
-	return sb.String(), nil
-}
-
-func (c *AnthropicClient) CreateChatCompletionStream(
-	ctx context.Context,
-	params model.RequestParameters,
-	messages []model.Message,
-	callback provider.ReplyCallback,
-	output chan<- string,
-) (string, error) {
-	request := buildRequest(params, messages)
-	request.Stream = true
-
-	resp, err := sendRequest(ctx, c, request)
-	if err != nil {
-		return "", err
-	}
-	defer resp.Body.Close()
-
-	scanner := bufio.NewScanner(resp.Body)
-	sb := strings.Builder{}
-
-	isToolCall := false
-
-	for scanner.Scan() {
-		line := scanner.Text()
-		line = strings.TrimSpace(line)
-
-		if len(line) == 0 {
-			continue
-		}
-
-		if line[0] == '{' {
-			var event map[string]interface{}
-			err := json.Unmarshal([]byte(line), &event)
-			if err != nil {
-				return "", fmt.Errorf("failed to unmarshal event data '%s': %v", line, err)
-			}
-			eventType, ok := event["type"].(string)
-			if !ok {
-				return "", fmt.Errorf("invalid event: %s", line)
-			}
-			switch eventType {
-			case "error":
-				return sb.String(), fmt.Errorf("an error occurred: %s", event["error"])
-			default:
-				return sb.String(), fmt.Errorf("unknown event type: %s", eventType)
-			}
-		} else if strings.HasPrefix(line, "data:") {
-			data := strings.TrimSpace(strings.TrimPrefix(line, "data:"))
-			var event map[string]interface{}
-			err := json.Unmarshal([]byte(data), &event)
-			if err != nil {
-				return "", fmt.Errorf("failed to unmarshal event data: %v", err)
-			}
-
-			eventType, ok := event["type"].(string)
-			if !ok {
-				return "", fmt.Errorf("invalid event type")
-			}
-
-			switch eventType {
-			case "message_start":
-				// noop
-			case "ping":
-				// write an empty string to signal start of text
-				output <- ""
-			case "content_block_start":
-				// ignore?
-			case "content_block_delta":
-				delta, ok := event["delta"].(map[string]interface{})
-				if !ok {
-					return "", fmt.Errorf("invalid content block delta")
-				}
-				text, ok := delta["text"].(string)
-				if !ok {
-					return "", fmt.Errorf("invalid text delta")
-				}
-				sb.WriteString(text)
-				output <- text
-			case "content_block_stop":
-				// ignore?
-			case "message_delta":
-				delta, ok := event["delta"].(map[string]interface{})
-				if !ok {
-					return "", fmt.Errorf("invalid message delta")
-				}
-				stopReason, ok := delta["stop_reason"].(string)
-				if ok && stopReason == "stop_sequence" {
-					stopSequence, ok := delta["stop_sequence"].(string)
-					if ok && stopSequence == FUNCTION_STOP_SEQUENCE {
-						content := sb.String()
-
-						start := strings.Index(content, "<function_calls>")
-						if start == -1 {
-							return content, fmt.Errorf("reached </function_calls> stop sequence but no opening tag found")
-						}
-
-						isToolCall = true
-
-						funcCallXml := content[start:]
-						funcCallXml += FUNCTION_STOP_SEQUENCE
-
-						sb.WriteString(FUNCTION_STOP_SEQUENCE)
-						output <- FUNCTION_STOP_SEQUENCE
-
-						// Extract function calls
-						var functionCalls XMLFunctionCalls
-						err := xml.Unmarshal([]byte(sb.String()), &functionCalls)
-						if err != nil {
-							return "", fmt.Errorf("failed to unmarshal function_calls: %v", err)
-						}
-
-						// Execute function calls
-						toolCall := model.Message{
-							Role: model.MessageRoleToolCall,
-							// xml stripped from content
-							Content:   content[:start],
-							ToolCalls: convertXMLFunctionCallsToToolCalls(functionCalls),
-						}
-
-						toolResults, err := tools.ExecuteToolCalls(toolCall.ToolCalls, params.ToolBag)
-						if err != nil {
-							return "", err
-						}
-
-						toolReply := model.Message{
-							Role:        model.MessageRoleToolResult,
-							ToolResults: toolResults,
-						}
-
-						if callback != nil {
-							callback(toolCall)
-							callback(toolReply)
-						}
-
-						// Recurse into CreateChatCompletionStream with the tool call replies
-						// added to the original messages
-						messages = append(append(messages, toolCall), toolReply)
-						return c.CreateChatCompletionStream(ctx, params, messages, callback, output)
-					}
-				}
-			case "message_stop":
-				// return the completed message
-				if callback != nil {
-					if !isToolCall {
-						callback(model.Message{
-							Role:    model.MessageRoleAssistant,
-							Content: sb.String(),
-						})
-					}
-				}
-				return sb.String(), nil
-			case "error":
-				return sb.String(), fmt.Errorf("an error occurred: %s", event["error"])
-			default:
-				fmt.Printf("\nUnrecognized event: %s\n", data)
-			}
-		}
-	}
-
-	if err := scanner.Err(); err != nil {
-		return "", fmt.Errorf("failed to read response body: %v", err)
-	}
-
-	return "", fmt.Errorf("unexpected end of stream")
-}
--- a/pkg/lmcli/provider/anthropic/tools.go
+++ b/pkg/lmcli/provider/anthropic/tools.go
@ -1,230 +0,0 @@
-package anthropic
-
-import (
-	"bytes"
-	"fmt"
-	"strings"
-	"text/template"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-)
-
-const TOOL_PREAMBLE = `You have access to the following tools when replying.
-
-You may call them like this:
-
-<function_calls>
-<invoke>
-<tool_name>$TOOL_NAME</tool_name>
-<parameters>
-<$PARAMETER_NAME>$PARAMETER_VALUE</$PARAMETER_NAME>
-...
-</parameters>
-</invoke>
-</function_calls>
-
-Here are the tools available:`
-
-const TOOL_PREAMBLE_FOOTER = `Recognize the utility of these tools in a broad range of different applications, and the power they give you to solve a wide range of different problems. However, ensure that the tools are used judiciously and only when clearly relevant to the user's request. Specifically:
-
-1. Only use a tool if the user has explicitly requested or provided information that warrants its use. Do not make assumptions about files or data existing without the user mentioning them.
-
-2. If there is ambiguity about whether using a tool is appropriate, ask a clarifying question to the user before proceeding. Confirm your understanding of their request and intent.
-
-3. Prioritize providing direct responses and explanations based on your own knowledge and understanding. Use tools to supplement and enhance your responses when clearly applicable, but not as a default action.`
-
-type XMLTools struct {
-	XMLName          struct{}             `xml:"tools"`
-	ToolDescriptions []XMLToolDescription `xml:"tool_description"`
-}
-
-type XMLToolDescription struct {
-	ToolName    string             `xml:"tool_name"`
-	Description string             `xml:"description"`
-	Parameters  []XMLToolParameter `xml:"parameters>parameter"`
-}
-
-type XMLToolParameter struct {
-	Name        string `xml:"name"`
-	Type        string `xml:"type"`
-	Description string `xml:"description"`
-}
-
-type XMLFunctionCalls struct {
-	XMLName struct{}            `xml:"function_calls"`
-	Invoke  []XMLFunctionInvoke `xml:"invoke"`
-}
-
-type XMLFunctionInvoke struct {
-	ToolName   string                      `xml:"tool_name"`
-	Parameters XMLFunctionInvokeParameters `xml:"parameters"`
-}
-
-type XMLFunctionInvokeParameters struct {
-	String string `xml:",innerxml"`
-}
-
-type XMLFunctionResults struct {
-	XMLName struct{}            `xml:"function_results"`
-	Result  []XMLFunctionResult `xml:"result"`
-}
-
-type XMLFunctionResult struct {
-	ToolName string `xml:"tool_name"`
-	Stdout   string `xml:"stdout"`
-}
-
-// accepts raw XML from XMLFunctionInvokeParameters.String, returns map of
-// parameters name to value
-func parseFunctionParametersXML(params string) map[string]interface{} {
-	lines := strings.Split(params, "\n")
-	ret := make(map[string]interface{}, len(lines))
-	for _, line := range lines {
-		i := strings.Index(line, ">")
-		if i == -1 {
-			continue
-		}
-		j := strings.Index(line, "</")
-		if j == -1 {
-			continue
-		}
-		// chop from after opening < to first > to get parameter name,
-		// then chop after > to first </ to get parameter value
-		ret[line[1:i]] = line[i+1 : j]
-	}
-	return ret
-}
-
-func convertToolsToXMLTools(tools []model.Tool) XMLTools {
-	converted := make([]XMLToolDescription, len(tools))
-	for i, tool := range tools {
-		converted[i].ToolName = tool.Name
-		converted[i].Description = tool.Description
-
-		params := make([]XMLToolParameter, len(tool.Parameters))
-		for j, param := range tool.Parameters {
-			params[j].Name = param.Name
-			params[j].Description = param.Description
-			params[j].Type = param.Type
-		}
-
-		converted[i].Parameters = params
-	}
-	return XMLTools{
-		ToolDescriptions: converted,
-	}
-}
-
-func convertXMLFunctionCallsToToolCalls(functionCalls XMLFunctionCalls) []model.ToolCall {
-	toolCalls := make([]model.ToolCall, len(functionCalls.Invoke))
-	for i, invoke := range functionCalls.Invoke {
-		toolCalls[i].Name = invoke.ToolName
-		toolCalls[i].Parameters = parseFunctionParametersXML(invoke.Parameters.String)
-	}
-	return toolCalls
-}
-
-func convertToolCallsToXMLFunctionCalls(toolCalls []model.ToolCall) XMLFunctionCalls {
-	converted := make([]XMLFunctionInvoke, len(toolCalls))
-	for i, toolCall := range toolCalls {
-		var params XMLFunctionInvokeParameters
-		var paramXML string
-		for key, value := range toolCall.Parameters {
-			paramXML += fmt.Sprintf("<%s>%v</%s>\n", key, value, key)
-		}
-		params.String = paramXML
-		converted[i] = XMLFunctionInvoke{
-			ToolName:   toolCall.Name,
-			Parameters: params,
-		}
-	}
-	return XMLFunctionCalls{
-		Invoke: converted,
-	}
-}
-
-func convertToolResultsToXMLFunctionResult(toolResults []model.ToolResult) XMLFunctionResults {
-	converted := make([]XMLFunctionResult, len(toolResults))
-	for i, result := range toolResults {
-		converted[i].ToolName = result.ToolName
-		converted[i].Stdout = result.Result
-	}
-	return XMLFunctionResults{
-		Result: converted,
-	}
-}
-
-func buildToolsSystemPrompt(tools []model.Tool) string {
-	xmlTools := convertToolsToXMLTools(tools)
-	xmlToolsString, err := xmlTools.XMLString()
-	if err != nil {
-		panic("Could not serialize []model.Tool to XMLTools")
-	}
-	return TOOL_PREAMBLE + "\n\n" + xmlToolsString + "\n\n" + TOOL_PREAMBLE_FOOTER
-}
-
-func (x XMLTools) XMLString() (string, error) {
-	tmpl, err := template.New("tools").Parse(`<tools>
-{{range .ToolDescriptions}}<tool_description>
-<tool_name>{{.ToolName}}</tool_name>
-<description>
-{{.Description}}
-</description>
-<parameters>
-{{range .Parameters}}<parameter>
-<name>{{.Name}}</name>
-<type>{{.Type}}</type>
-<description>{{.Description}}</description>
-</parameter>
-{{end}}</parameters>
-</tool_description>
-{{end}}</tools>`)
-	if err != nil {
-		return "", err
-	}
-
-	var buf bytes.Buffer
-	if err := tmpl.Execute(&buf, x); err != nil {
-		return "", err
-	}
-
-	return buf.String(), nil
-}
-
-func (x XMLFunctionResults) XMLString() (string, error) {
-	tmpl, err := template.New("function_results").Parse(`<function_results>
-{{range .Result}}<result>
-<tool_name>{{.ToolName}}</tool_name>
-<stdout>{{.Stdout}}</stdout>
-</result>
-{{end}}</function_results>`)
-	if err != nil {
-		return "", err
-	}
-
-	var buf bytes.Buffer
-	if err := tmpl.Execute(&buf, x); err != nil {
-		return "", err
-	}
-
-	return buf.String(), nil
-}
-
-func (x XMLFunctionCalls) XMLString() (string, error) {
-	tmpl, err := template.New("function_calls").Parse(`<function_calls>
-{{range .Invoke}}<invoke>
-<tool_name>{{.ToolName}}</tool_name>
-<parameters>{{.Parameters.String}}</parameters>
-</invoke>
-{{end}}</function_calls>`)
-	if err != nil {
-		return "", err
-	}
-
-	var buf bytes.Buffer
-	if err := tmpl.Execute(&buf, x); err != nil {
-		return "", err
-	}
-
-	return buf.String(), nil
-}
--- a/pkg/lmcli/provider/openai/openai.go
+++ b/pkg/lmcli/provider/openai/openai.go
@ -1,278 +0,0 @@
-package openai
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"io"
-	"strings"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/provider"
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/tools"
-	openai "github.com/sashabaranov/go-openai"
-)
-
-type OpenAIClient struct {
-	APIKey string
-}
-
-type OpenAIToolParameters struct {
-	Type       string                         `json:"type"`
-	Properties map[string]OpenAIToolParameter `json:"properties,omitempty"`
-	Required   []string                       `json:"required,omitempty"`
-}
-
-type OpenAIToolParameter struct {
-	Type        string   `json:"type"`
-	Description string   `json:"description"`
-	Enum        []string `json:"enum,omitempty"`
-}
-
-func convertTools(tools []model.Tool) []openai.Tool {
-	openaiTools := make([]openai.Tool, len(tools))
-	for i, tool := range tools {
-		openaiTools[i].Type = "function"
-
-		params := make(map[string]OpenAIToolParameter)
-		var required []string
-
-		for _, param := range tool.Parameters {
-			params[param.Name] = OpenAIToolParameter{
-				Type:        param.Type,
-				Description: param.Description,
-				Enum:        param.Enum,
-			}
-			if param.Required {
-				required = append(required, param.Name)
-			}
-		}
-
-		openaiTools[i].Function = openai.FunctionDefinition{
-			Name:        tool.Name,
-			Description: tool.Description,
-			Parameters: OpenAIToolParameters{
-				Type:       "object",
-				Properties: params,
-				Required:   required,
-			},
-		}
-	}
-	return openaiTools
-}
-
-func convertToolCallToOpenAI(toolCalls []model.ToolCall) []openai.ToolCall {
-	converted := make([]openai.ToolCall, len(toolCalls))
-	for i, call := range toolCalls {
-		converted[i].Type = "function"
-		converted[i].ID = call.ID
-		converted[i].Function.Name = call.Name
-
-		json, _ := json.Marshal(call.Parameters)
-		converted[i].Function.Arguments = string(json)
-	}
-	return converted
-}
-
-func convertToolCallToAPI(toolCalls []openai.ToolCall) []model.ToolCall {
-	converted := make([]model.ToolCall, len(toolCalls))
-	for i, call := range toolCalls {
-		converted[i].ID = call.ID
-		converted[i].Name = call.Function.Name
-		json.Unmarshal([]byte(call.Function.Arguments), &converted[i].Parameters)
-	}
-	return converted
-}
-
-func createChatCompletionRequest(
-	c *OpenAIClient,
-	params model.RequestParameters,
-	messages []model.Message,
-) openai.ChatCompletionRequest {
-	requestMessages := make([]openai.ChatCompletionMessage, 0, len(messages))
-
-	for _, m := range messages {
-		switch m.Role {
-		case "tool_call":
-			message := openai.ChatCompletionMessage{}
-			message.Role = "assistant"
-			message.Content = m.Content
-			message.ToolCalls = convertToolCallToOpenAI(m.ToolCalls)
-			requestMessages = append(requestMessages, message)
-		case "tool_result":
-			// expand tool_result messages' results into multiple openAI messages
-			for _, result := range m.ToolResults {
-				message := openai.ChatCompletionMessage{}
-				message.Role = "tool"
-				message.Content = result.Result
-				message.ToolCallID = result.ToolCallID
-				requestMessages = append(requestMessages, message)
-			}
-		default:
-			message := openai.ChatCompletionMessage{}
-			message.Role = string(m.Role)
-			message.Content = m.Content
-			requestMessages = append(requestMessages, message)
-		}
-	}
-
-	request := openai.ChatCompletionRequest{
-		Model:       params.Model,
-		MaxTokens:   params.MaxTokens,
-		Temperature: params.Temperature,
-		Messages:    requestMessages,
-		N:           1, // limit responses to 1 "choice". we use choices[0] to reference it
-	}
-
-	if len(params.ToolBag) > 0 {
-		request.Tools = convertTools(params.ToolBag)
-		request.ToolChoice = "auto"
-	}
-
-	return request
-}
-
-func handleToolCalls(
-	params model.RequestParameters,
-	content string,
-	toolCalls []openai.ToolCall,
-) ([]model.Message, error) {
-	toolCall := model.Message{
-		Role:      model.MessageRoleToolCall,
-		Content:   content,
-		ToolCalls: convertToolCallToAPI(toolCalls),
-	}
-
-	toolResults, err := tools.ExecuteToolCalls(toolCall.ToolCalls, params.ToolBag)
-	if err != nil {
-		return nil, err
-	}
-
-	toolResult := model.Message{
-		Role:        model.MessageRoleToolResult,
-		ToolResults: toolResults,
-	}
-
-	return []model.Message{toolCall, toolResult}, nil
-}
-
-func (c *OpenAIClient) CreateChatCompletion(
-	ctx context.Context,
-	params model.RequestParameters,
-	messages []model.Message,
-	callback provider.ReplyCallback,
-) (string, error) {
-	client := openai.NewClient(c.APIKey)
-	req := createChatCompletionRequest(c, params, messages)
-	resp, err := client.CreateChatCompletion(ctx, req)
-	if err != nil {
-		return "", err
-	}
-
-	choice := resp.Choices[0]
-
-	toolCalls := choice.Message.ToolCalls
-	if len(toolCalls) > 0 {
-		results, err := handleToolCalls(params, choice.Message.Content, toolCalls)
-		if err != nil {
-			return "", err
-		}
-		if callback != nil {
-			for _, result := range results {
-				callback(result)
-			}
-		}
-
-		// Recurse into CreateChatCompletion with the tool call replies
-		messages = append(messages, results...)
-		return c.CreateChatCompletion(ctx, params, messages, callback)
-	}
-
-	if callback != nil {
-		callback(model.Message{
-			Role:    model.MessageRoleAssistant,
-			Content: choice.Message.Content,
-		})
-	}
-
-	// Return the user-facing message.
-	return choice.Message.Content, nil
-}
-
-func (c *OpenAIClient) CreateChatCompletionStream(
-	ctx context.Context,
-	params model.RequestParameters,
-	messages []model.Message,
-	callback provider.ReplyCallback,
-	output chan<- string,
-) (string, error) {
-	client := openai.NewClient(c.APIKey)
-	req := createChatCompletionRequest(c, params, messages)
-
-	stream, err := client.CreateChatCompletionStream(ctx, req)
-	if err != nil {
-		return "", err
-	}
-	defer stream.Close()
-
-	content := strings.Builder{}
-	toolCalls := []openai.ToolCall{}
-
-	// Iterate stream segments
-	for {
-		response, e := stream.Recv()
-		if errors.Is(e, io.EOF) {
-			break
-		}
-
-		if e != nil {
-			err = e
-			break
-		}
-
-		delta := response.Choices[0].Delta
-		if len(delta.ToolCalls) > 0 {
-			// Construct streamed tool_call arguments
-			for _, tc := range delta.ToolCalls {
-				if tc.Index == nil {
-					return "", fmt.Errorf("Unexpected nil index for streamed tool call.")
-				}
-				if len(toolCalls) <= *tc.Index {
-					toolCalls = append(toolCalls, tc)
-				} else {
-					toolCalls[*tc.Index].Function.Arguments += tc.Function.Arguments
-				}
-			}
-		} else {
-			output <- delta.Content
-			content.WriteString(delta.Content)
-		}
-	}
-
-	if len(toolCalls) > 0 {
-		results, err := handleToolCalls(params, content.String(), toolCalls)
-		if err != nil {
-			return content.String(), err
-		}
-
-		if callback != nil {
-			for _, result := range results {
-				callback(result)
-			}
-		}
-
-		// Recurse into CreateChatCompletionStream with the tool call replies
-		messages = append(messages, results...)
-		return c.CreateChatCompletionStream(ctx, params, messages, callback, output)
-	} else {
-		if callback != nil {
-			callback(model.Message{
-				Role:    model.MessageRoleAssistant,
-				Content: content.String(),
-			})
-		}
-	}
-
-	return content.String(), err
-}
--- a/pkg/lmcli/provider/provider.go
+++ b/pkg/lmcli/provider/provider.go
@ -1,31 +0,0 @@
-package provider
-
-import (
-	"context"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-)
-
-type ReplyCallback func(model.Message)
-
-type ChatCompletionClient interface {
-	// CreateChatCompletion requests a response to the provided messages.
-	// Replies are appended to the given replies struct, and the
-	// complete user-facing response is returned as a string.
-	CreateChatCompletion(
-		ctx context.Context,
-		params model.RequestParameters,
-		messages []model.Message,
-		callback ReplyCallback,
-	) (string, error)
-
-	// Like CreateChageCompletion, except the response is streamed via
-	// the output channel as it's received.
-	CreateChatCompletionStream(
-		ctx context.Context,
-		params model.RequestParameters,
-		messages []model.Message,
-		callback ReplyCallback,
-		output chan<- string,
-	) (string, error)
-}
--- a/pkg/lmcli/store.go
+++ b/pkg/lmcli/store.go
@ -1,132 +0,0 @@
-package lmcli
-
-import (
-	"database/sql"
-	"errors"
-	"fmt"
-	"strings"
-	"time"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	sqids "github.com/sqids/sqids-go"
-	"gorm.io/gorm"
-)
-
-type ConversationStore interface {
-	Conversations() ([]model.Conversation, error)
-
-	ConversationByShortName(shortName string) (*model.Conversation, error)
-	ConversationShortNameCompletions(search string) []string
-
-	SaveConversation(conversation *model.Conversation) error
-	DeleteConversation(conversation *model.Conversation) error
-
-	Messages(conversation *model.Conversation) ([]model.Message, error)
-	LastMessage(conversation *model.Conversation) (*model.Message, error)
-
-	SaveMessage(message *model.Message) error
-	DeleteMessage(message *model.Message) error
-	UpdateMessage(message *model.Message) error
-	AddReply(conversation *model.Conversation, message model.Message) (*model.Message, error)
-}
-
-type SQLStore struct {
-	db    *gorm.DB
-	sqids *sqids.Sqids
-}
-
-func NewSQLStore(db *gorm.DB) (*SQLStore, error) {
-	models := []any{
-		&model.Conversation{},
-		&model.Message{},
-	}
-
-	for _, x := range models {
-		err := db.AutoMigrate(x)
-		if err != nil {
-			return nil, fmt.Errorf("Could not perform database migrations: %v", err)
-		}
-	}
-
-	_sqids, _ := sqids.New(sqids.Options{MinLength: 4})
-	return &SQLStore{db, _sqids}, nil
-}
-
-func (s *SQLStore) SaveConversation(conversation *model.Conversation) error {
-	err := s.db.Save(&conversation).Error
-	if err != nil {
-		return err
-	}
-
-	if !conversation.ShortName.Valid {
-		shortName, _ := s.sqids.Encode([]uint64{uint64(conversation.ID)})
-		conversation.ShortName = sql.NullString{String: shortName, Valid: true}
-		err = s.db.Save(&conversation).Error
-	}
-
-	return err
-}
-
-func (s *SQLStore) DeleteConversation(conversation *model.Conversation) error {
-	s.db.Where("conversation_id = ?", conversation.ID).Delete(&model.Message{})
-	return s.db.Delete(&conversation).Error
-}
-
-func (s *SQLStore) SaveMessage(message *model.Message) error {
-	return s.db.Create(message).Error
-}
-
-func (s *SQLStore) DeleteMessage(message *model.Message) error {
-	return s.db.Delete(&message).Error
-}
-
-func (s *SQLStore) UpdateMessage(message *model.Message) error {
-	return s.db.Updates(&message).Error
-}
-
-func (s *SQLStore) Conversations() ([]model.Conversation, error) {
-	var conversations []model.Conversation
-	err := s.db.Find(&conversations).Error
-	return conversations, err
-}
-
-func (s *SQLStore) ConversationShortNameCompletions(shortName string) []string {
-	var completions []string
-	conversations, _ := s.Conversations() // ignore error for completions
-	for _, conversation := range conversations {
-		if shortName == "" || strings.HasPrefix(conversation.ShortName.String, shortName) {
-			completions = append(completions, fmt.Sprintf("%s\t%s", conversation.ShortName.String, conversation.Title))
-		}
-	}
-	return completions
-}
-
-func (s *SQLStore) ConversationByShortName(shortName string) (*model.Conversation, error) {
-	if shortName == "" {
-		return nil, errors.New("shortName is empty")
-	}
-	var conversation model.Conversation
-	err := s.db.Where("short_name = ?", shortName).Find(&conversation).Error
-	return &conversation, err
-}
-
-func (s *SQLStore) Messages(conversation *model.Conversation) ([]model.Message, error) {
-	var messages []model.Message
-	err := s.db.Where("conversation_id = ?", conversation.ID).Find(&messages).Error
-	return messages, err
-}
-
-func (s *SQLStore) LastMessage(conversation *model.Conversation) (*model.Message, error) {
-	var message model.Message
-	err := s.db.Where("conversation_id = ?", conversation.ID).Last(&message).Error
-	return &message, err
-}
-
-// AddReply adds the given messages as a reply to the given conversation, can be
-// used to easily copy a message associated with one conversation, to another
-func (s *SQLStore) AddReply(c *model.Conversation, m model.Message) (*model.Message, error) {
-	m.ConversationID = c.ID
-	m.ID = 0
-	m.CreatedAt = time.Time{}
-	return &m, s.SaveMessage(&m)
-}
--- a/pkg/lmcli/tools/file_replace_lines.go
+++ b/pkg/lmcli/tools/file_replace_lines.go
@ -1,133 +0,0 @@
-package tools
-
-import (
-	"fmt"
-	"os"
-	"strings"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	toolutil "git.mlow.ca/mlow/lmcli/pkg/lmcli/tools/util"
-)
-
-const FILE_REPLACE_LINES_DESCRIPTION = `Replace or remove a range of lines within a file, must specify path.
-
-Useful for re-writing snippets/blocks of code or entire functions.
-
-Plan your edits carefully and ensure any new content matches the flow and indentation of surrounding text.`
-
-var FileReplaceLinesTool = model.Tool{
-	Name:        "file_replace_lines",
-	Description: FILE_REPLACE_LINES_DESCRIPTION,
-	Parameters: []model.ToolParameter{
-		{
-			Name:        "path",
-			Type:        "string",
-			Description: "Path of the file to be modified, relative to the current working directory.",
-			Required:    true,
-		},
-		{
-			Name:        "start_line",
-			Type:        "integer",
-			Description: `Line number which specifies the start of the replacement range (inclusive).`,
-			Required:    true,
-		},
-		{
-			Name:        "end_line",
-			Type:        "integer",
-			Description: `Line number which specifies the end of the replacement range (inclusive). If unset, range extends to end of file.`,
-		},
-		{
-			Name:        "content",
-			Type:        "string",
-			Description: `Content to replace specified range. Omit to remove the specified range.`,
-		},
-	},
-	Impl: func(tool *model.Tool, args map[string]interface{}) (string, error) {
-		tmp, ok := args["path"]
-		if !ok {
-			return "", fmt.Errorf("path parameter to write_file was not included.")
-		}
-		path, ok := tmp.(string)
-		if !ok {
-			return "", fmt.Errorf("Invalid path in function arguments: %v", tmp)
-		}
-		var start_line int
-		tmp, ok = args["start_line"]
-		if ok {
-			tmp, ok := tmp.(float64)
-			if !ok {
-				return "", fmt.Errorf("Invalid start_line in function arguments: %v", tmp)
-			}
-			start_line = int(tmp)
-		}
-		var end_line int
-		tmp, ok = args["end_line"]
-		if ok {
-			tmp, ok := tmp.(float64)
-			if !ok {
-				return "", fmt.Errorf("Invalid end_line in function arguments: %v", tmp)
-			}
-			end_line = int(tmp)
-		}
-		var content string
-		tmp, ok = args["content"]
-		if ok {
-			content, ok = tmp.(string)
-			if !ok {
-				return "", fmt.Errorf("Invalid content in function arguments: %v", tmp)
-			}
-		}
-
-		result := fileReplaceLines(path, start_line, end_line, content)
-		ret, err := result.ToJson()
-		if err != nil {
-			return "", fmt.Errorf("Could not serialize result: %v", err)
-		}
-		return ret, nil
-	},
-}
-
-func fileReplaceLines(path string, startLine int, endLine int, content string) model.CallResult {
-	ok, reason := toolutil.IsPathWithinCWD(path)
-	if !ok {
-		return model.CallResult{Message: reason}
-	}
-
-	// Read the existing file's content
-	data, err := os.ReadFile(path)
-	if err != nil {
-		if !os.IsNotExist(err) {
-			return model.CallResult{Message: fmt.Sprintf("Could not read path: %s", err.Error())}
-		}
-		_, err = os.Create(path)
-		if err != nil {
-			return model.CallResult{Message: fmt.Sprintf("Could not create new file: %s", err.Error())}
-		}
-		data = []byte{}
-	}
-
-	if startLine < 1 {
-		return model.CallResult{Message: "start_line cannot be less than 1"}
-	}
-
-	lines := strings.Split(string(data), "\n")
-	contentLines := strings.Split(strings.Trim(content, "\n"), "\n")
-
-	if endLine == 0 || endLine > len(lines) {
-		endLine = len(lines)
-	}
-
-	before := lines[:startLine-1]
-	after := lines[endLine:]
-
-	lines = append(before, append(contentLines, after...)...)
-	newContent := strings.Join(lines, "\n")
-
-	// Join the lines and write back to the file
-	err = os.WriteFile(path, []byte(newContent), 0644)
-	if err != nil {
-		return model.CallResult{Message: fmt.Sprintf("Could not write to path: %s", err.Error())}
-	}
-
-	return model.CallResult{Result: newContent}
-}
--- a/pkg/lmcli/tools/tools.go
+++ b/pkg/lmcli/tools/tools.go
@ -1,47 +0,0 @@
-package tools
-
-import (
-	"fmt"
-
-	"git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-)
-
-var AvailableTools map[string]model.Tool = map[string]model.Tool{
-	"read_dir":           ReadDirTool,
-	"read_file":          ReadFileTool,
-	"write_file":         WriteFileTool,
-	"file_insert_lines":  FileInsertLinesTool,
-	"file_replace_lines": FileReplaceLinesTool,
-}
-
-func ExecuteToolCalls(toolCalls []model.ToolCall, toolBag []model.Tool) ([]model.ToolResult, error) {
-	var toolResults []model.ToolResult
-	for _, toolCall := range toolCalls {
-		var tool *model.Tool
-		for _, available := range toolBag {
-			if available.Name == toolCall.Name {
-				tool = &available
-				break
-			}
-		}
-		if tool == nil {
-			return nil, fmt.Errorf("Requested tool '%s' does not exist. Hallucination?", toolCall.Name)
-		}
-
-		// Execute the tool
-		result, err := tool.Impl(tool, toolCall.Parameters)
-		if err != nil {
-			// This can happen if the model missed or supplied invalid tool args
-			return nil, fmt.Errorf("Tool '%s' error: %v\n", toolCall.Name, err)
-		}
-
-		toolResult := model.ToolResult{
-			ToolCallID: toolCall.ID,
-			ToolName: toolCall.Name,
-			Result: result,
-		}
-
-		toolResults = append(toolResults, toolResult)
-	}
-	return toolResults, nil
-}
--- a/pkg/provider/anthropic/anthropic.go
+++ b/pkg/provider/anthropic/anthropic.go
@ -0,0 +1,447 @@
+package anthropic
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+)
+
+const ANTHROPIC_VERSION = "2023-06-01"
+
+type AnthropicClient struct {
+	APIKey  string
+	BaseURL string
+}
+
+type ChatCompletionMessage struct {
+	Role    string      `json:"role"`
+	Content interface{} `json:"content"`
+}
+
+type Tool struct {
+	Name        string      `json:"name"`
+	Description string      `json:"description"`
+	InputSchema InputSchema `json:"input_schema"`
+}
+
+type InputSchema struct {
+	Type       string              `json:"type"`
+	Properties map[string]Property `json:"properties"`
+	Required   []string            `json:"required"`
+}
+
+type Property struct {
+	Type        string   `json:"type"`
+	Description string   `json:"description"`
+	Enum        []string `json:"enum,omitempty"`
+}
+
+type ChatCompletionRequest struct {
+	Model       string                  `json:"model"`
+	Messages    []ChatCompletionMessage `json:"messages"`
+	System      string                  `json:"system,omitempty"`
+	Tools       []Tool                  `json:"tools,omitempty"`
+	MaxTokens   int                     `json:"max_tokens"`
+	Temperature float32                 `json:"temperature,omitempty"`
+	Stream      bool                    `json:"stream"`
+}
+
+type ContentBlock struct {
+	Type                   string      `json:"type"`
+	Text                   string      `json:"text,omitempty"`
+	ID                     string      `json:"id,omitempty"`
+	Name                   string      `json:"name,omitempty"`
+	Input                  interface{} `json:"input,omitempty"`
+	partialJsonAccumulator string
+}
+
+type ChatCompletionResponse struct {
+	ID         string         `json:"id"`
+	Type       string         `json:"type"`
+	Role       string         `json:"role"`
+	Model      string         `json:"model"`
+	Content    []ContentBlock `json:"content"`
+	StopReason string         `json:"stop_reason"`
+	Usage      Usage          `json:"usage"`
+}
+
+type Usage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+}
+
+type StreamEvent struct {
+	Type    string      `json:"type"`
+	Message interface{} `json:"message,omitempty"`
+	Index   int         `json:"index,omitempty"`
+	Delta   interface{} `json:"delta,omitempty"`
+}
+
+func convertTools(tools []api.ToolSpec) []Tool {
+	anthropicTools := make([]Tool, len(tools))
+	for i, tool := range tools {
+		properties := make(map[string]Property)
+		for _, param := range tool.Parameters {
+			properties[param.Name] = Property{
+				Type:        param.Type,
+				Description: param.Description,
+				Enum:        param.Enum,
+			}
+		}
+
+		var required []string
+		for _, param := range tool.Parameters {
+			if param.Required {
+				required = append(required, param.Name)
+			}
+		}
+
+		anthropicTools[i] = Tool{
+			Name:        tool.Name,
+			Description: tool.Description,
+			InputSchema: InputSchema{
+				Type:       "object",
+				Properties: properties,
+				Required:   required,
+			},
+		}
+	}
+	return anthropicTools
+}
+
+func createChatCompletionRequest(
+	params provider.RequestParameters,
+	messages []api.Message,
+) (string, ChatCompletionRequest) {
+	requestMessages := make([]ChatCompletionMessage, 0, len(messages))
+	var systemMessage string
+
+	for _, m := range messages {
+		if m.Role == api.MessageRoleSystem {
+			systemMessage = m.Content
+			continue
+		}
+
+		var content interface{}
+		role := string(m.Role)
+
+		switch m.Role {
+		case api.MessageRoleToolCall:
+			role = "assistant"
+			contentBlocks := make([]map[string]interface{}, 0)
+			if m.Content != "" {
+				contentBlocks = append(contentBlocks, map[string]interface{}{
+					"type": "text",
+					"text": m.Content,
+				})
+			}
+			for _, toolCall := range m.ToolCalls {
+				contentBlocks = append(contentBlocks, map[string]interface{}{
+					"type":  "tool_use",
+					"id":    toolCall.ID,
+					"name":  toolCall.Name,
+					"input": toolCall.Parameters,
+				})
+			}
+			content = contentBlocks
+
+		case api.MessageRoleToolResult:
+			role = "user"
+			contentBlocks := make([]map[string]interface{}, 0)
+			for _, result := range m.ToolResults {
+				contentBlock := map[string]interface{}{
+					"type":        "tool_result",
+					"tool_use_id": result.ToolCallID,
+					"content":     result.Result,
+				}
+				contentBlocks = append(contentBlocks, contentBlock)
+			}
+			content = contentBlocks
+
+		default:
+			content = m.Content
+		}
+
+		requestMessages = append(requestMessages, ChatCompletionMessage{
+			Role:    role,
+			Content: content,
+		})
+	}
+
+	request := ChatCompletionRequest{
+		Model:       params.Model,
+		Messages:    requestMessages,
+		System:      systemMessage,
+		MaxTokens:   params.MaxTokens,
+		Temperature: params.Temperature,
+	}
+
+	if len(params.Toolbox) > 0 {
+		request.Tools = convertTools(params.Toolbox)
+	}
+
+	var prefill string
+	if len(messages) > 0 && messages[len(messages)-1].Role == api.MessageRoleAssistant {
+		// Prompting on an assitant message, use its content as prefill
+		prefill = messages[len(messages)-1].Content
+	}
+
+	return prefill, request
+}
+
+func (c *AnthropicClient) sendRequest(ctx context.Context, r ChatCompletionRequest) (*http.Response, error) {
+	jsonData, err := json.Marshal(r)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", c.BaseURL+"/v1/messages", bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create HTTP request: %w", err)
+	}
+
+	req.Header.Set("x-api-key", c.APIKey)
+	req.Header.Set("anthropic-version", ANTHROPIC_VERSION)
+	req.Header.Set("content-type", "application/json")
+
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+
+	if resp.StatusCode != 200 {
+		bytes, _ := io.ReadAll(resp.Body)
+		return resp, fmt.Errorf("%v", string(bytes))
+	}
+
+	return resp, err
+}
+
+func (c *AnthropicClient) CreateChatCompletion(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("can't create completion from no messages")
+	}
+
+	_, req := createChatCompletionRequest(params, messages)
+	req.Stream = false
+
+	resp, err := c.sendRequest(ctx, req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to send request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	var completionResp ChatCompletionResponse
+	err = json.NewDecoder(resp.Body).Decode(&completionResp)
+	if err != nil {
+		return nil, fmt.Errorf("failed to decode response: %w", err)
+	}
+
+	return convertResponseToMessage(completionResp)
+}
+
+func (c *AnthropicClient) CreateChatCompletionStream(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+	output chan<- provider.Chunk,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("can't create completion from no messages")
+	}
+
+	prefill, req := createChatCompletionRequest(params, messages)
+	req.Stream = true
+
+	resp, err := c.sendRequest(ctx, req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to send request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	contentBlocks := make(map[int]*ContentBlock)
+	var finalMessage *ChatCompletionResponse
+
+	var firstChunkReceived bool
+
+	reader := bufio.NewReader(resp.Body)
+	for {
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		default:
+			line, err := reader.ReadBytes('\n')
+			if err != nil {
+				if err == io.EOF {
+					break
+				}
+				return nil, fmt.Errorf("error reading stream: %w", err)
+			}
+
+			line = bytes.TrimSpace(line)
+			if len(line) == 0 || !bytes.HasPrefix(line, []byte("data: ")) {
+				continue
+			}
+
+			line = bytes.TrimPrefix(line, []byte("data: "))
+
+			var streamEvent StreamEvent
+			err = json.Unmarshal(line, &streamEvent)
+			if err != nil {
+				return nil, fmt.Errorf("failed to unmarshal stream event: %w", err)
+			}
+
+			switch streamEvent.Type {
+			case "message_start":
+				finalMessage = &ChatCompletionResponse{}
+				err = json.Unmarshal(line, &struct {
+					Message *ChatCompletionResponse `json:"message"`
+				}{Message: finalMessage})
+				if err != nil {
+					return nil, fmt.Errorf("failed to unmarshal message_start: %w", err)
+				}
+			case "content_block_start":
+				var contentBlockStart struct {
+					Index        int          `json:"index"`
+					ContentBlock ContentBlock `json:"content_block"`
+				}
+				err = json.Unmarshal(line, &contentBlockStart)
+				if err != nil {
+					return nil, fmt.Errorf("failed to unmarshal content_block_start: %w", err)
+				}
+
+				contentBlocks[contentBlockStart.Index] = &contentBlockStart.ContentBlock
+			case "content_block_delta":
+				if streamEvent.Index >= len(contentBlocks) {
+					return nil, fmt.Errorf("received delta for non-existent content block index: %d", streamEvent.Index)
+				}
+
+				block := contentBlocks[streamEvent.Index]
+				delta, ok := streamEvent.Delta.(map[string]interface{})
+				if !ok {
+					return nil, fmt.Errorf("unexpected delta type: %T", streamEvent.Delta)
+				}
+
+				deltaType, ok := delta["type"].(string)
+				if !ok {
+					return nil, fmt.Errorf("delta missing type field")
+				}
+
+				switch deltaType {
+				case "text_delta":
+					if text, ok := delta["text"].(string); ok {
+						if !firstChunkReceived {
+							if prefill == "" {
+								// if there is no prefil, ensure we trim leading whitespace
+								text = strings.TrimSpace(text)
+							}
+							firstChunkReceived = true
+						}
+						block.Text += text
+						output <- provider.Chunk{
+							Content:    text,
+							// rough, anthropic performs some chunking
+							TokenCount: uint(len(strings.Split(text, " "))),
+						}
+					}
+				case "input_json_delta":
+					if block.Type != "tool_use" {
+						return nil, fmt.Errorf("received input_json_delta for non-tool_use block")
+					}
+					if partialJSON, ok := delta["partial_json"].(string); ok {
+						block.partialJsonAccumulator += partialJSON
+					}
+				}
+
+			case "content_block_stop":
+				if streamEvent.Index >= len(contentBlocks) {
+					return nil, fmt.Errorf("received stop for non-existent content block index: %d", streamEvent.Index)
+				}
+
+				block := contentBlocks[streamEvent.Index]
+				if block.Type == "tool_use" && block.partialJsonAccumulator != "" {
+					var inputData map[string]interface{}
+					err := json.Unmarshal([]byte(block.partialJsonAccumulator), &inputData)
+					if err != nil {
+						return nil, fmt.Errorf("failed to unmarshal accumulated JSON for tool use: %w", err)
+					}
+					block.Input = inputData
+				}
+			case "message_delta":
+				if finalMessage == nil {
+					return nil, fmt.Errorf("received message_delta before message_start")
+				}
+				delta, ok := streamEvent.Delta.(map[string]interface{})
+				if !ok {
+					return nil, fmt.Errorf("unexpected delta type in message_delta: %T", streamEvent.Delta)
+				}
+				if stopReason, ok := delta["stop_reason"].(string); ok {
+					finalMessage.StopReason = stopReason
+				}
+
+			case "message_stop":
+				// End of the stream
+				goto END_STREAM
+
+			case "error":
+				return nil, fmt.Errorf("received error event: %v", streamEvent.Message)
+
+			default:
+				// Ignore unknown event types
+			}
+		}
+	}
+
+END_STREAM:
+	if finalMessage == nil {
+		return nil, fmt.Errorf("no final message received")
+	}
+
+	finalMessage.Content = make([]ContentBlock, len(contentBlocks))
+	for _, v := range contentBlocks {
+		finalMessage.Content = append(finalMessage.Content, *v)
+	}
+
+	return convertResponseToMessage(*finalMessage)
+}
+
+func convertResponseToMessage(resp ChatCompletionResponse) (*api.Message, error) {
+	content := strings.Builder{}
+	var toolCalls []api.ToolCall
+
+	for _, block := range resp.Content {
+		switch block.Type {
+		case "text":
+			content.WriteString(block.Text)
+		case "tool_use":
+			parameters, ok := block.Input.(map[string]interface{})
+			if !ok {
+				return nil, fmt.Errorf("unexpected type for tool call parameters: %T", block.Input)
+			}
+			toolCalls = append(toolCalls, api.ToolCall{
+				ID:         block.ID,
+				Name:       block.Name,
+				Parameters: parameters,
+			})
+		}
+	}
+
+	if len(toolCalls) > 0 {
+		return api.NewMessageWithToolCalls(content.String(), toolCalls), nil
+	}
+
+	return api.NewMessageWithAssistant(content.String()), nil
+}
--- a/pkg/provider/google/google.go
+++ b/pkg/provider/google/google.go
@ -0,0 +1,436 @@
+package google
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+)
+
+type Client struct {
+	APIKey  string
+	BaseURL string
+}
+
+type ContentPart struct {
+	Text         string            `json:"text,omitempty"`
+	FunctionCall *FunctionCall     `json:"functionCall,omitempty"`
+	FunctionResp *FunctionResponse `json:"functionResponse,omitempty"`
+}
+
+type FunctionCall struct {
+	Name string            `json:"name"`
+	Args map[string]string `json:"args"`
+}
+
+type FunctionResponse struct {
+	Name     string      `json:"name"`
+	Response interface{} `json:"response"`
+}
+
+type Content struct {
+	Role  string        `json:"role"`
+	Parts []ContentPart `json:"parts"`
+}
+
+type GenerationConfig struct {
+	MaxOutputTokens *int     `json:"maxOutputTokens,omitempty"`
+	Temperature     *float32 `json:"temperature,omitempty"`
+	TopP            *float32 `json:"topP,omitempty"`
+	TopK            *int     `json:"topK,omitempty"`
+}
+
+type GenerateContentRequest struct {
+	Contents          []Content         `json:"contents"`
+	Tools             []Tool            `json:"tools,omitempty"`
+	SystemInstruction *Content          `json:"systemInstruction,omitempty"`
+	GenerationConfig  *GenerationConfig `json:"generationConfig,omitempty"`
+}
+
+type Candidate struct {
+	Content      Content `json:"content"`
+	FinishReason string  `json:"finishReason"`
+	Index        int     `json:"index"`
+}
+
+type UsageMetadata struct {
+	PromptTokenCount     int `json:"promptTokenCount"`
+	CandidatesTokenCount int `json:"candidatesTokenCount"`
+	TotalTokenCount      int `json:"totalTokenCount"`
+}
+
+type GenerateContentResponse struct {
+	Candidates    []Candidate   `json:"candidates"`
+	UsageMetadata UsageMetadata `json:"usageMetadata"`
+}
+
+type Tool struct {
+	FunctionDeclarations []FunctionDeclaration `json:"functionDeclarations"`
+}
+
+type FunctionDeclaration struct {
+	Name        string         `json:"name"`
+	Description string         `json:"description"`
+	Parameters  ToolParameters `json:"parameters"`
+}
+
+type ToolParameters struct {
+	Type       string                   `json:"type"`
+	Properties map[string]ToolParameter `json:"properties,omitempty"`
+	Required   []string                 `json:"required,omitempty"`
+}
+
+type ToolParameter struct {
+	Type        string   `json:"type"`
+	Description string   `json:"description"`
+	Values      []string `json:"values,omitempty"`
+}
+
+func convertTools(tools []api.ToolSpec) []Tool {
+	geminiTools := make([]Tool, len(tools))
+	for i, tool := range tools {
+		params := make(map[string]ToolParameter)
+		var required []string
+
+		for _, param := range tool.Parameters {
+			// TODO: proper enum handing
+			params[param.Name] = ToolParameter{
+				Type:        param.Type,
+				Description: param.Description,
+				Values:      param.Enum,
+			}
+			if param.Required {
+				required = append(required, param.Name)
+			}
+		}
+
+		geminiTools[i] = Tool{
+			FunctionDeclarations: []FunctionDeclaration{
+				{
+					Name:        tool.Name,
+					Description: tool.Description,
+					Parameters: ToolParameters{
+						Type:       "OBJECT",
+						Properties: params,
+						Required:   required,
+					},
+				},
+			},
+		}
+	}
+	return geminiTools
+}
+
+func convertToolCallToGemini(toolCalls []api.ToolCall) []ContentPart {
+	converted := make([]ContentPart, len(toolCalls))
+	for i, call := range toolCalls {
+		args := make(map[string]string)
+		for k, v := range call.Parameters {
+			args[k] = fmt.Sprintf("%v", v)
+		}
+		converted[i].FunctionCall = &FunctionCall{
+			Name: call.Name,
+			Args: args,
+		}
+	}
+	return converted
+}
+
+func convertToolCallToAPI(functionCalls []FunctionCall) []api.ToolCall {
+	converted := make([]api.ToolCall, len(functionCalls))
+	for i, call := range functionCalls {
+		params := make(map[string]interface{})
+		for k, v := range call.Args {
+			params[k] = v
+		}
+		converted[i].Name = call.Name
+		converted[i].Parameters = params
+	}
+	return converted
+}
+
+func convertToolResultsToGemini(toolResults []api.ToolResult) ([]FunctionResponse, error) {
+	results := make([]FunctionResponse, len(toolResults))
+	for i, result := range toolResults {
+		var obj interface{}
+		err := json.Unmarshal([]byte(result.Result), &obj)
+		if err != nil {
+			return nil, fmt.Errorf("Could not unmarshal %s: %v", result.Result, err)
+		}
+		results[i] = FunctionResponse{
+			Name:     result.ToolName,
+			Response: obj,
+		}
+	}
+	return results, nil
+}
+
+func createGenerateContentRequest(
+	params provider.RequestParameters,
+	messages []api.Message,
+) (*GenerateContentRequest, error) {
+	requestContents := make([]Content, 0, len(messages))
+
+	startIdx := 0
+	var system string
+	if len(messages) > 0 && messages[0].Role == api.MessageRoleSystem {
+		system = messages[0].Content
+		startIdx = 1
+	}
+
+	for _, m := range messages[startIdx:] {
+		switch m.Role {
+		case "tool_call":
+			content := Content{
+				Role:  "model",
+				Parts: convertToolCallToGemini(m.ToolCalls),
+			}
+			requestContents = append(requestContents, content)
+		case "tool_result":
+			results, err := convertToolResultsToGemini(m.ToolResults)
+			if err != nil {
+				return nil, err
+			}
+			// expand tool_result messages' results into multiple gemini messages
+			for _, result := range results {
+				content := Content{
+					Role: "function",
+					Parts: []ContentPart{
+						{
+							FunctionResp: &result,
+						},
+					},
+				}
+				requestContents = append(requestContents, content)
+			}
+		default:
+			var role string
+			switch m.Role {
+			case api.MessageRoleAssistant:
+				role = "model"
+			case api.MessageRoleUser:
+				role = "user"
+			}
+
+			if role == "" {
+				panic("Unhandled role: " + m.Role)
+			}
+
+			content := Content{
+				Role: role,
+				Parts: []ContentPart{
+					{
+						Text: m.Content,
+					},
+				},
+			}
+			requestContents = append(requestContents, content)
+		}
+	}
+
+	request := &GenerateContentRequest{
+		Contents: requestContents,
+		GenerationConfig: &GenerationConfig{
+			MaxOutputTokens: &params.MaxTokens,
+			Temperature:     &params.Temperature,
+			TopP:            &params.TopP,
+		},
+	}
+
+	if system != "" {
+		request.SystemInstruction = &Content{
+			Parts: []ContentPart{
+				{
+					Text: system,
+				},
+			},
+		}
+	}
+
+	if len(params.Toolbox) > 0 {
+		request.Tools = convertTools(params.Toolbox)
+	}
+
+	return request, nil
+}
+
+func (c *Client) sendRequest(req *http.Request) (*http.Response, error) {
+	req.Header.Set("Content-Type", "application/json")
+
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+
+	if resp.StatusCode != 200 {
+		bytes, _ := io.ReadAll(resp.Body)
+		return resp, fmt.Errorf("%v", string(bytes))
+	}
+
+	return resp, err
+}
+
+func (c *Client) CreateChatCompletion(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("Can't create completion from no messages")
+	}
+
+	req, err := createGenerateContentRequest(params, messages)
+	if err != nil {
+		return nil, err
+	}
+	jsonData, err := json.Marshal(req)
+	if err != nil {
+		return nil, err
+	}
+
+	url := fmt.Sprintf(
+		"%s/v1beta/models/%s:generateContent?key=%s",
+		c.BaseURL, params.Model, c.APIKey,
+	)
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, err
+	}
+
+	resp, err := c.sendRequest(httpReq)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	var completionResp GenerateContentResponse
+	err = json.NewDecoder(resp.Body).Decode(&completionResp)
+	if err != nil {
+		return nil, err
+	}
+
+	choice := completionResp.Candidates[0]
+
+	var content string
+	lastMessage := messages[len(messages)-1]
+	if lastMessage.Role.IsAssistant() {
+		content = lastMessage.Content
+	}
+
+	var toolCalls []FunctionCall
+	for _, part := range choice.Content.Parts {
+		if part.Text != "" {
+			content += part.Text
+		}
+
+		if part.FunctionCall != nil {
+			toolCalls = append(toolCalls, *part.FunctionCall)
+		}
+	}
+
+	if len(toolCalls) > 0 {
+		return api.NewMessageWithToolCalls(content, convertToolCallToAPI(toolCalls)), nil
+	}
+
+	return api.NewMessageWithAssistant(content), nil
+}
+
+func (c *Client) CreateChatCompletionStream(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+	output chan<- provider.Chunk,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("Can't create completion from no messages")
+	}
+
+	req, err := createGenerateContentRequest(params, messages)
+	if err != nil {
+		return nil, err
+	}
+	jsonData, err := json.Marshal(req)
+	if err != nil {
+		return nil, err
+	}
+
+	url := fmt.Sprintf(
+		"%s/v1beta/models/%s:streamGenerateContent?key=%s&alt=sse",
+		c.BaseURL, params.Model, c.APIKey,
+	)
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, err
+	}
+
+	resp, err := c.sendRequest(httpReq)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	content := strings.Builder{}
+
+	lastMessage := messages[len(messages)-1]
+	if lastMessage.Role.IsAssistant() {
+		content.WriteString(lastMessage.Content)
+	}
+
+	var toolCalls []FunctionCall
+
+	reader := bufio.NewReader(resp.Body)
+
+	lastTokenCount := 0
+	for {
+		line, err := reader.ReadBytes('\n')
+		if err != nil {
+			if err == io.EOF {
+				break
+			}
+			return nil, err
+		}
+
+		line = bytes.TrimSpace(line)
+		if len(line) == 0 || !bytes.HasPrefix(line, []byte("data: ")) {
+			continue
+		}
+
+		line = bytes.TrimPrefix(line, []byte("data: "))
+
+		var resp GenerateContentResponse
+		err = json.Unmarshal(line, &resp)
+		if err != nil {
+			return nil, err
+		}
+
+		tokens := resp.UsageMetadata.CandidatesTokenCount - lastTokenCount
+		lastTokenCount += tokens
+
+		choice := resp.Candidates[0]
+		for _, part := range choice.Content.Parts {
+			if part.FunctionCall != nil {
+				toolCalls = append(toolCalls, *part.FunctionCall)
+			} else if part.Text != "" {
+				output <- provider.Chunk{
+					Content:    part.Text,
+					TokenCount: uint(tokens),
+				}
+				content.WriteString(part.Text)
+			}
+		}
+	}
+
+	if len(toolCalls) > 0 {
+		return api.NewMessageWithToolCalls(content.String(), convertToolCallToAPI(toolCalls)), nil
+	}
+
+	return api.NewMessageWithAssistant(content.String()), nil
+}
--- a/pkg/provider/ollama/ollama.go
+++ b/pkg/provider/ollama/ollama.go
@ -0,0 +1,183 @@
+package ollama
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+)
+
+type OllamaClient struct {
+	BaseURL string
+}
+
+type OllamaMessage struct {
+	Role    string `json:"role"`
+	Content string `json:"content"`
+}
+
+type OllamaRequest struct {
+	Model    string          `json:"model"`
+	Messages []OllamaMessage `json:"messages"`
+	Stream   bool            `json:"stream"`
+}
+
+type OllamaResponse struct {
+	Model              string        `json:"model"`
+	CreatedAt          string        `json:"created_at"`
+	Message            OllamaMessage `json:"message"`
+	Done               bool          `json:"done"`
+	TotalDuration      uint64        `json:"total_duration,omitempty"`
+	LoadDuration       uint64        `json:"load_duration,omitempty"`
+	PromptEvalCount    uint64        `json:"prompt_eval_count,omitempty"`
+	PromptEvalDuration uint64        `json:"prompt_eval_duration,omitempty"`
+	EvalCount          uint64        `json:"eval_count,omitempty"`
+	EvalDuration       uint64        `json:"eval_duration,omitempty"`
+}
+
+func createOllamaRequest(
+	params provider.RequestParameters,
+	messages []api.Message,
+) OllamaRequest {
+	requestMessages := make([]OllamaMessage, 0, len(messages))
+
+	for _, m := range messages {
+		message := OllamaMessage{
+			Role:    string(m.Role),
+			Content: m.Content,
+		}
+		requestMessages = append(requestMessages, message)
+	}
+
+	request := OllamaRequest{
+		Model:    params.Model,
+		Messages: requestMessages,
+	}
+
+	return request
+}
+
+func (c *OllamaClient) sendRequest(req *http.Request) (*http.Response, error) {
+	req.Header.Set("Content-Type", "application/json")
+
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+
+	if resp.StatusCode != 200 {
+		bytes, _ := io.ReadAll(resp.Body)
+		return resp, fmt.Errorf("%v", string(bytes))
+	}
+
+	return resp, nil
+}
+
+func (c *OllamaClient) CreateChatCompletion(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("Can't create completion from no messages")
+	}
+
+	req := createOllamaRequest(params, messages)
+	req.Stream = false
+
+	jsonData, err := json.Marshal(req)
+	if err != nil {
+		return nil, err
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", c.BaseURL+"/chat", bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, err
+	}
+
+	resp, err := c.sendRequest(httpReq)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	var completionResp OllamaResponse
+	err = json.NewDecoder(resp.Body).Decode(&completionResp)
+	if err != nil {
+		return nil, err
+	}
+
+	return api.NewMessageWithAssistant(completionResp.Message.Content), nil
+}
+
+func (c *OllamaClient) CreateChatCompletionStream(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+	output chan<- provider.Chunk,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("Can't create completion from no messages")
+	}
+
+	req := createOllamaRequest(params, messages)
+	req.Stream = true
+
+	jsonData, err := json.Marshal(req)
+	if err != nil {
+		return nil, err
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", c.BaseURL+"/chat", bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, err
+	}
+
+	resp, err := c.sendRequest(httpReq)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	content := strings.Builder{}
+
+	reader := bufio.NewReader(resp.Body)
+	for {
+		line, err := reader.ReadBytes('\n')
+		if err != nil {
+			if err == io.EOF {
+				break
+			}
+			return nil, err
+		}
+
+		line = bytes.TrimSpace(line)
+		if len(line) == 0 {
+			continue
+		}
+
+		var streamResp OllamaResponse
+		err = json.Unmarshal(line, &streamResp)
+		if err != nil {
+			return nil, err
+		}
+
+		if len(streamResp.Message.Content) > 0 {
+			output <- provider.Chunk{
+				Content:    streamResp.Message.Content,
+				TokenCount: 1,
+			}
+			content.WriteString(streamResp.Message.Content)
+		}
+	}
+
+	return api.NewMessageWithAssistant(content.String()), nil
+}
--- a/pkg/provider/openai/openai.go
+++ b/pkg/provider/openai/openai.go
@ -0,0 +1,343 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+)
+
+type OpenAIClient struct {
+	APIKey  string
+	BaseURL string
+	Headers map[string]string
+}
+
+type ChatCompletionMessage struct {
+	Role       string     `json:"role"`
+	Content    string     `json:"content,omitempty"`
+	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`
+	ToolCallID string     `json:"tool_call_id,omitempty"`
+}
+
+type ToolCall struct {
+	Type     string             `json:"type"`
+	ID       string             `json:"id"`
+	Index    *int               `json:"index,omitempty"`
+	Function FunctionDefinition `json:"function"`
+}
+
+type FunctionDefinition struct {
+	Name        string         `json:"name"`
+	Description string         `json:"description"`
+	Parameters  ToolParameters `json:"parameters"`
+	Arguments   string         `json:"arguments,omitempty"`
+}
+
+type ToolParameters struct {
+	Type       string                   `json:"type"`
+	Properties map[string]ToolParameter `json:"properties,omitempty"`
+	Required   []string                 `json:"required,omitempty"`
+}
+
+type ToolParameter struct {
+	Type        string   `json:"type"`
+	Description string   `json:"description"`
+	Enum        []string `json:"enum,omitempty"`
+}
+
+type Tool struct {
+	Type     string             `json:"type"`
+	Function FunctionDefinition `json:"function"`
+}
+
+type ChatCompletionRequest struct {
+	Model       string                  `json:"model"`
+	MaxTokens   int                     `json:"max_tokens,omitempty"`
+	Temperature float32                 `json:"temperature,omitempty"`
+	Messages    []ChatCompletionMessage `json:"messages"`
+	N           int                     `json:"n"`
+	Tools       []Tool                  `json:"tools,omitempty"`
+	ToolChoice  string                  `json:"tool_choice,omitempty"`
+	Stream      bool                    `json:"stream,omitempty"`
+}
+
+type ChatCompletionChoice struct {
+	Message ChatCompletionMessage `json:"message"`
+}
+
+type ChatCompletionResponse struct {
+	Choices []ChatCompletionChoice `json:"choices"`
+}
+
+type ChatCompletionStreamChoice struct {
+	Delta ChatCompletionMessage `json:"delta"`
+}
+
+type ChatCompletionStreamResponse struct {
+	Choices []ChatCompletionStreamChoice `json:"choices"`
+}
+
+func convertTools(tools []api.ToolSpec) []Tool {
+	openaiTools := make([]Tool, len(tools))
+	for i, tool := range tools {
+		openaiTools[i].Type = "function"
+
+		params := make(map[string]ToolParameter)
+		var required []string
+
+		for _, param := range tool.Parameters {
+			params[param.Name] = ToolParameter{
+				Type:        param.Type,
+				Description: param.Description,
+				Enum:        param.Enum,
+			}
+			if param.Required {
+				required = append(required, param.Name)
+			}
+		}
+
+		openaiTools[i].Function = FunctionDefinition{
+			Name:        tool.Name,
+			Description: tool.Description,
+			Parameters: ToolParameters{
+				Type:       "object",
+				Properties: params,
+				Required:   required,
+			},
+		}
+	}
+	return openaiTools
+}
+
+func convertToolCallToOpenAI(toolCalls []api.ToolCall) []ToolCall {
+	converted := make([]ToolCall, len(toolCalls))
+	for i, call := range toolCalls {
+		converted[i].Type = "function"
+		converted[i].ID = call.ID
+		converted[i].Function.Name = call.Name
+
+		json, _ := json.Marshal(call.Parameters)
+		converted[i].Function.Arguments = string(json)
+	}
+	return converted
+}
+
+func convertToolCallToAPI(toolCalls []ToolCall) []api.ToolCall {
+	converted := make([]api.ToolCall, len(toolCalls))
+	for i, call := range toolCalls {
+		converted[i].ID = call.ID
+		converted[i].Name = call.Function.Name
+		json.Unmarshal([]byte(call.Function.Arguments), &converted[i].Parameters)
+	}
+	return converted
+}
+
+func createChatCompletionRequest(
+	params provider.RequestParameters,
+	messages []api.Message,
+) ChatCompletionRequest {
+	requestMessages := make([]ChatCompletionMessage, 0, len(messages))
+
+	for _, m := range messages {
+		switch m.Role {
+		case "tool_call":
+			message := ChatCompletionMessage{}
+			message.Role = "assistant"
+			message.Content = m.Content
+			message.ToolCalls = convertToolCallToOpenAI(m.ToolCalls)
+			requestMessages = append(requestMessages, message)
+		case "tool_result":
+			// expand tool_result messages' results into multiple openAI messages
+			for _, result := range m.ToolResults {
+				message := ChatCompletionMessage{}
+				message.Role = "tool"
+				message.Content = result.Result
+				message.ToolCallID = result.ToolCallID
+				requestMessages = append(requestMessages, message)
+			}
+		default:
+			message := ChatCompletionMessage{}
+			message.Role = string(m.Role)
+			message.Content = m.Content
+			requestMessages = append(requestMessages, message)
+		}
+	}
+
+	request := ChatCompletionRequest{
+		Model:       params.Model,
+		MaxTokens:   params.MaxTokens,
+		Temperature: params.Temperature,
+		Messages:    requestMessages,
+		N:           1, // limit responses to 1 "choice". we use choices[0] to reference it
+	}
+
+	if len(params.Toolbox) > 0 {
+		request.Tools = convertTools(params.Toolbox)
+		request.ToolChoice = "auto"
+	}
+
+	return request
+}
+
+func (c *OpenAIClient) sendRequest(ctx context.Context, r ChatCompletionRequest) (*http.Response, error) {
+	jsonData, err := json.Marshal(r)
+	if err != nil {
+		return nil, err
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", c.BaseURL+"/v1/chat/completions", bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, err
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer "+c.APIKey)
+	for header, val := range c.Headers {
+		req.Header.Set(header, val)
+	}
+
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+
+	if resp.StatusCode != 200 {
+		bytes, _ := io.ReadAll(resp.Body)
+		return resp, fmt.Errorf("%v", string(bytes))
+	}
+
+	return resp, err
+}
+
+func (c *OpenAIClient) CreateChatCompletion(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("Can't create completion from no messages")
+	}
+
+	req := createChatCompletionRequest(params, messages)
+
+	resp, err := c.sendRequest(ctx, req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	var completionResp ChatCompletionResponse
+	err = json.NewDecoder(resp.Body).Decode(&completionResp)
+	if err != nil {
+		return nil, err
+	}
+
+	choice := completionResp.Choices[0]
+
+	var content string
+	lastMessage := messages[len(messages)-1]
+	if lastMessage.Role.IsAssistant() {
+		content = lastMessage.Content + choice.Message.Content
+	} else {
+		content = choice.Message.Content
+	}
+
+	toolCalls := choice.Message.ToolCalls
+	if len(toolCalls) > 0 {
+		return api.NewMessageWithToolCalls(content, convertToolCallToAPI(toolCalls)), nil
+	}
+
+	return api.NewMessageWithAssistant(content), nil
+}
+
+func (c *OpenAIClient) CreateChatCompletionStream(
+	ctx context.Context,
+	params provider.RequestParameters,
+	messages []api.Message,
+	output chan<- provider.Chunk,
+) (*api.Message, error) {
+	if len(messages) == 0 {
+		return nil, fmt.Errorf("Can't create completion from no messages")
+	}
+
+	req := createChatCompletionRequest(params, messages)
+	req.Stream = true
+
+	resp, err := c.sendRequest(ctx, req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	content := strings.Builder{}
+	toolCalls := []ToolCall{}
+
+	lastMessage := messages[len(messages)-1]
+	if lastMessage.Role.IsAssistant() {
+		content.WriteString(lastMessage.Content)
+	}
+
+	reader := bufio.NewReader(resp.Body)
+	for {
+		line, err := reader.ReadBytes('\n')
+		if err != nil {
+			if err == io.EOF {
+				break
+			}
+			return nil, err
+		}
+
+		line = bytes.TrimSpace(line)
+		if len(line) == 0 || !bytes.HasPrefix(line, []byte("data: ")) {
+			continue
+		}
+
+		line = bytes.TrimPrefix(line, []byte("data: "))
+		if bytes.Equal(line, []byte("[DONE]")) {
+			break
+		}
+
+		var streamResp ChatCompletionStreamResponse
+		err = json.Unmarshal(line, &streamResp)
+		if err != nil {
+			return nil, err
+		}
+
+		delta := streamResp.Choices[0].Delta
+		if len(delta.ToolCalls) > 0 {
+			// Construct streamed tool_call arguments
+			for _, tc := range delta.ToolCalls {
+				if tc.Index == nil {
+					return nil, fmt.Errorf("Unexpected nil index for streamed tool call.")
+				}
+				if len(toolCalls) <= *tc.Index {
+					toolCalls = append(toolCalls, tc)
+				} else {
+					toolCalls[*tc.Index].Function.Arguments += tc.Function.Arguments
+				}
+			}
+		}
+		if len(delta.Content) > 0 {
+			output <- provider.Chunk{
+				Content:    delta.Content,
+				TokenCount: 1,
+			}
+			content.WriteString(delta.Content)
+		}
+	}
+
+	if len(toolCalls) > 0 {
+		return api.NewMessageWithToolCalls(content.String(), convertToolCallToAPI(toolCalls)), nil
+	}
+
+	return api.NewMessageWithAssistant(content.String()), nil
+}
--- a/pkg/provider/provider.go
+++ b/pkg/provider/provider.go
@ -0,0 +1,41 @@
+package provider
+
+import (
+	"context"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+)
+
+type Chunk struct {
+	Content    string
+	TokenCount uint
+}
+
+type RequestParameters struct {
+	Model string
+
+	MaxTokens   int
+	Temperature float32
+	TopP        float32
+
+	Toolbox []api.ToolSpec
+}
+
+type ChatCompletionProvider interface {
+	// CreateChatCompletion generates a chat completion response to the
+	// provided messages.
+	CreateChatCompletion(
+		ctx context.Context,
+		params RequestParameters,
+		messages []api.Message,
+	) (*api.Message, error)
+
+	// Like CreateChageCompletion, except the response is streamed via
+	// the output channel.
+	CreateChatCompletionStream(
+		ctx context.Context,
+		params RequestParameters,
+		messages []api.Message,
+		chunks chan<- Chunk,
+	) (*api.Message, error)
+}
--- a/pkg/tui/bubbles/confirmprompt.go
+++ b/pkg/tui/bubbles/confirmprompt.go
@ -0,0 +1,67 @@
+package bubbles
+
+import (
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+type ConfirmPrompt struct {
+	Question string
+	Style    lipgloss.Style
+	Payload  interface{}
+	value    bool
+	answered bool
+	focused  bool
+}
+
+func NewConfirmPrompt(question string, payload interface{}) ConfirmPrompt {
+	return ConfirmPrompt{
+		Question: question,
+		Style:    lipgloss.NewStyle(),
+		Payload:  payload,
+		focused:  true, // focus by default
+	}
+}
+
+type MsgConfirmPromptAnswered struct {
+	Value   bool
+	Payload interface{}
+}
+
+func (b ConfirmPrompt) Update(msg tea.Msg) (ConfirmPrompt, tea.Cmd) {
+	switch msg := msg.(type) {
+	case tea.KeyMsg:
+		if !b.focused || b.answered {
+			return b, nil
+		}
+		switch msg.String() {
+		case "y", "Y":
+			b.value = true
+			b.answered = true
+			b.focused = false
+			return b, func() tea.Msg { return MsgConfirmPromptAnswered{true, b.Payload} }
+		case "n", "N", "esc":
+			b.value = false
+			b.answered = true
+			b.focused = false
+			return b, func() tea.Msg { return MsgConfirmPromptAnswered{false, b.Payload} }
+		}
+	}
+	return b, nil
+}
+
+func (b ConfirmPrompt) View() string {
+	return b.Style.Render(b.Question) + lipgloss.NewStyle().Faint(true).Render(" (y/n)")
+}
+
+func (b *ConfirmPrompt) Focus() {
+	b.focused = true
+}
+
+func (b *ConfirmPrompt) Blur() {
+	b.focused = false
+}
+
+func (b ConfirmPrompt) Focused() bool {
+	return b.focused
+}
--- a/pkg/tui/bubbles/list/list.go
+++ b/pkg/tui/bubbles/list/list.go
@ -0,0 +1,260 @@
+package list
+
+import (
+	"fmt"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	tuiutil "git.mlow.ca/mlow/lmcli/pkg/tui/util"
+	"github.com/charmbracelet/bubbles/textinput"
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+type Option struct {
+	Label string
+	Value interface{}
+}
+
+type OptionGroup struct {
+	Name    string
+	Options []Option
+}
+
+type Model struct {
+	ID            int
+	HeaderStyle   lipgloss.Style
+	ItemStyle     lipgloss.Style
+	SelectedStyle lipgloss.Style
+	ItemRender    func(Option, bool) string
+
+	Width  int
+	Height int
+
+	optionGroups    []OptionGroup
+	selected        int
+	filterInput     textinput.Model
+	filteredIndices []filteredIndex
+	content         viewport.Model
+	itemYOffsets    []int
+}
+
+type filteredIndex struct {
+	groupIndex  int
+	optionIndex int
+}
+
+type MsgOptionSelected struct {
+	ID     int
+	Option Option
+}
+
+func New(opts []Option) Model {
+	return NewWithGroups([]OptionGroup{{Options: opts}})
+}
+
+func NewWithGroups(groups []OptionGroup) Model {
+	ti := textinput.New()
+	ti.Prompt = "/"
+	ti.PromptStyle = lipgloss.NewStyle().Faint(true)
+
+	m := Model{
+		HeaderStyle:   lipgloss.NewStyle().Bold(true).Foreground(lipgloss.Color("12")).Padding(1, 0, 1, 1),
+		ItemStyle:     lipgloss.NewStyle(),
+		SelectedStyle: lipgloss.NewStyle().Faint(true).Foreground(lipgloss.Color("6")),
+
+		optionGroups:    groups,
+		selected:        0,
+		filterInput:     ti,
+		filteredIndices: make([]filteredIndex, 0),
+		content:         viewport.New(0, 0),
+		itemYOffsets:    make([]int, 0),
+	}
+
+	m.filterItems()
+	m.content.SetContent(m.renderOptionsList())
+	return m
+}
+
+func (m *Model) Focused() {
+	m.filterInput.Focused()
+}
+
+func (m *Model) Focus() {
+	m.filterInput.Focus()
+}
+
+func (m *Model) Blur() {
+	m.filterInput.Blur()
+}
+
+func (m *Model) filterItems() {
+	filterText := strings.ToLower(m.filterInput.Value())
+
+	var prevSelection *filteredIndex
+	if m.selected <= len(m.filteredIndices)-1 {
+		prevSelection = &m.filteredIndices[m.selected]
+	}
+
+	m.filteredIndices = make([]filteredIndex, 0)
+
+	for groupIndex, group := range m.optionGroups {
+		for optionIndex, option := range group.Options {
+			if filterText == "" ||
+				strings.Contains(strings.ToLower(option.Label), filterText) ||
+				(group.Name != "" && strings.Contains(strings.ToLower(group.Name), filterText)) {
+				m.filteredIndices = append(m.filteredIndices, filteredIndex{groupIndex, optionIndex})
+			}
+		}
+	}
+
+	found := false
+	if len(m.filteredIndices) > 0 && prevSelection != nil {
+		// Preserve previous selection if possible
+		for i, filterIdx := range m.filteredIndices {
+			if prevSelection.groupIndex == filterIdx.groupIndex && prevSelection.optionIndex == filterIdx.optionIndex {
+				m.selected = i
+				found = true
+				break
+			}
+		}
+	}
+	if !found {
+		m.selected = 0
+	}
+}
+
+func (m *Model) Update(msg tea.Msg) (Model, tea.Cmd) {
+	var cmd tea.Cmd
+
+	switch msg := msg.(type) {
+	case tea.KeyMsg:
+		if m.filterInput.Focused() {
+			switch msg.String() {
+			case "esc":
+				m.filterInput.Blur()
+				m.filterInput.SetValue("")
+				m.filterItems()
+				m.refreshContent()
+				return *m, shared.KeyHandled(msg)
+			case "enter":
+				m.filterInput.Blur()
+				m.refreshContent()
+				break
+			case "up", "down":
+				break
+			default:
+				m.filterInput, cmd = m.filterInput.Update(msg)
+				m.filterItems()
+				m.refreshContent()
+				return *m, cmd
+			}
+		}
+
+		switch msg.String() {
+		case "up", "k":
+			m.moveSelection(-1)
+			return *m, shared.KeyHandled(msg)
+		case "down", "j":
+			m.moveSelection(1)
+			return *m, shared.KeyHandled(msg)
+		case "enter":
+			return *m, func() tea.Msg {
+				idx := m.filteredIndices[m.selected]
+				return MsgOptionSelected{
+					ID:     m.ID,
+					Option: m.optionGroups[idx.groupIndex].Options[idx.optionIndex],
+				}
+			}
+		case "/":
+			m.filterInput.Focus()
+			return *m, textinput.Blink
+		}
+	}
+
+	m.content, cmd = m.content.Update(msg)
+	return *m, cmd
+}
+
+func (m *Model) refreshContent() {
+	m.content.SetContent(m.renderOptionsList())
+	m.ensureSelectedVisible()
+}
+
+func (m *Model) ensureSelectedVisible() {
+	if m.selected == 0 {
+		m.content.GotoTop()
+	} else if m.selected == len(m.filteredIndices)-1 {
+		m.content.GotoBottom()
+	} else {
+		tuiutil.ScrollIntoView(&m.content, m.itemYOffsets[m.selected], 0)
+	}
+}
+
+func (m *Model) moveSelection(delta int) {
+	prev := m.selected
+	m.selected = min(len(m.filteredIndices)-1, max(0, m.selected+delta))
+	if prev != m.selected {
+		m.refreshContent()
+	}
+}
+
+func (m *Model) View() string {
+	filter := ""
+	if m.filterInput.Focused() {
+		m.filterInput.Width = m.Width
+		filter = m.filterInput.View()
+	}
+
+	contentHeight := m.Height - tuiutil.Height(filter)
+	m.content.Width, m.content.Height = m.Width, contentHeight
+
+	parts := []string{m.content.View()}
+	if filter != "" {
+		parts = append(parts, filter)
+	}
+	return lipgloss.JoinVertical(lipgloss.Left, parts...)
+}
+
+func (m *Model) renderOptionsList() string {
+	yOffset := 0
+	lastGroupIndex := -1
+	m.itemYOffsets = make([]int, len(m.filteredIndices))
+
+	var sb strings.Builder
+	for i, idx := range m.filteredIndices {
+		if idx.groupIndex != lastGroupIndex {
+			group := m.optionGroups[idx.groupIndex].Name
+			if group != "" {
+				headingStr := m.HeaderStyle.Render(group)
+				yOffset += tuiutil.Height(headingStr)
+				sb.WriteString(headingStr)
+				sb.WriteRune('\n')
+			}
+			lastGroupIndex = idx.groupIndex
+		}
+
+		m.itemYOffsets[i] = yOffset
+		option := m.optionGroups[idx.groupIndex].Options[idx.optionIndex]
+		var item string
+		if m.ItemRender != nil {
+			item = m.ItemRender(option, i == m.selected)
+		} else {
+			prefix := "  "
+			if i == m.selected {
+				prefix = "> "
+				item = m.SelectedStyle.Render(option.Label)
+			} else {
+				item = m.ItemStyle.Render(option.Label)
+			}
+			item = fmt.Sprintf("%s%s", prefix, item)
+		}
+		sb.WriteString(item)
+		yOffset += tuiutil.Height(item)
+		if i < len(m.filteredIndices)-1 {
+			sb.WriteRune('\n')
+		}
+	}
+	return sb.String()
+}
--- a/pkg/tui/model/model.go
+++ b/pkg/tui/model/model.go
@ -0,0 +1,281 @@
+package model
+
+import (
+	"context"
+	"fmt"
+
+	"git.mlow.ca/mlow/lmcli/pkg/agents"
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+	"github.com/charmbracelet/lipgloss"
+)
+
+type AppModel struct {
+	Ctx           *lmcli.Context
+	Conversations conversation.ConversationList
+	Conversation  conversation.Conversation
+	Messages      []conversation.Message
+	Model         string
+	ProviderName  string
+	Provider      provider.ChatCompletionProvider
+	Agent         *lmcli.Agent
+}
+
+func NewAppModel(ctx *lmcli.Context, initialConversation *conversation.Conversation) *AppModel {
+	app := &AppModel{
+		Ctx:          ctx,
+		Model:        *ctx.Config.Defaults.Model,
+	}
+
+	if initialConversation == nil {
+		app.NewConversation()
+	} else {
+
+	}
+
+	model, provider, _, _ := ctx.GetModelProvider(*ctx.Config.Defaults.Model, "")
+	app.Model = model
+	app.ProviderName = provider
+	app.Agent = ctx.GetAgent(ctx.Config.Defaults.Agent)
+	return app
+}
+
+var (
+	defaultStyle = lipgloss.NewStyle().Faint(true)
+	accentStyle  = defaultStyle.Foreground(lipgloss.Color("6"))
+)
+
+func (a *AppModel) ActiveModel(style lipgloss.Style) string {
+	defaultStyle := style.Inherit(defaultStyle)
+	accentStyle := style.Inherit(accentStyle)
+	return defaultStyle.Render(a.Model) + accentStyle.Render("@") + defaultStyle.Render(a.ProviderName)
+}
+
+type MessageCycleDirection int
+
+const (
+	CycleNext MessageCycleDirection = 1
+	CyclePrev MessageCycleDirection = -1
+)
+
+func (m *AppModel) ClearConversation() {
+	m.Conversation = conversation.Conversation{}
+	m.Messages = []conversation.Message{}
+}
+
+func (m *AppModel) ApplySystemPrompt() {
+	var system string
+	agent := m.Ctx.GetAgent(m.Ctx.Config.Defaults.Agent)
+	if agent != nil && agent.SystemPrompt != "" {
+		system = agent.SystemPrompt
+	}
+	if system == "" {
+		system = m.Ctx.DefaultSystemPrompt()
+	}
+	if system != "" {
+		m.Messages = conversation.ApplySystemPrompt(m.Messages, system, false)
+	}
+}
+
+func (m *AppModel) NewConversation() {
+	m.ClearConversation()
+	m.ApplySystemPrompt()
+}
+
+func (a *AppModel) LoadConversationMessages() ([]conversation.Message, error) {
+	messages, err := a.Ctx.Conversations.PathToLeaf(a.Conversation.SelectedRoot)
+	if err != nil {
+		return nil, fmt.Errorf("Could not load conversation messages: %v %v", a.Conversation.SelectedRoot, err)
+	}
+	return messages, nil
+}
+
+func (a *AppModel) GenerateConversationTitle(messages []conversation.Message) (string, error) {
+	return cmdutil.GenerateTitle(a.Ctx, messages)
+}
+
+func (a *AppModel) CloneMessage(message conversation.Message, selected bool) (*conversation.Message, error) {
+	msg, _, err := a.Ctx.Conversations.CloneBranch(message)
+	if err != nil {
+		return nil, fmt.Errorf("Could not clone message: %v", err)
+	}
+	if selected {
+		if msg.Parent == nil {
+			msg.Conversation.SelectedRoot = msg
+			err = a.Ctx.Conversations.UpdateConversation(msg.Conversation)
+		} else {
+			msg.Parent.SelectedReply = msg
+			err = a.Ctx.Conversations.UpdateMessage(msg.Parent)
+		}
+		if err != nil {
+			return nil, fmt.Errorf("Could not update selected message: %v", err)
+		}
+	}
+	return msg, nil
+}
+
+func (a *AppModel) UpdateMessageContent(message *conversation.Message) error {
+	return a.Ctx.Conversations.UpdateMessage(message)
+}
+
+func cycleSelectedMessage(selected *conversation.Message, choices []conversation.Message, dir MessageCycleDirection) (*conversation.Message, error) {
+	currentIndex := -1
+	for i, reply := range choices {
+		if reply.ID == selected.ID {
+			currentIndex = i
+			break
+		}
+	}
+
+	if currentIndex < 0 {
+		return nil, fmt.Errorf("Selected message %d not found in choices, this is a bug", selected.ID)
+	}
+
+	var next int
+	if dir == CyclePrev {
+		next = (currentIndex - 1 + len(choices)) % len(choices)
+	} else {
+		next = (currentIndex + 1) % len(choices)
+	}
+	return &choices[next], nil
+}
+
+func (a *AppModel) CycleSelectedRoot(conv *conversation.Conversation, dir MessageCycleDirection) (*conversation.Message, error) {
+	if len(conv.RootMessages) < 2 {
+		return nil, nil
+	}
+
+	nextRoot, err := cycleSelectedMessage(conv.SelectedRoot, conv.RootMessages, dir)
+	if err != nil {
+		return nil, err
+	}
+
+	conv.SelectedRoot = nextRoot
+	err = a.Ctx.Conversations.UpdateConversation(conv)
+	if err != nil {
+		return nil, fmt.Errorf("Could not update conversation SelectedRoot: %v", err)
+	}
+	return nextRoot, nil
+}
+
+func (a *AppModel) CycleSelectedReply(message *conversation.Message, dir MessageCycleDirection) (*conversation.Message, error) {
+	if len(message.Replies) < 2 {
+		return nil, nil
+	}
+
+	nextReply, err := cycleSelectedMessage(message.SelectedReply, message.Replies, dir)
+	if err != nil {
+		return nil, err
+	}
+
+	message.SelectedReply = nextReply
+	err = a.Ctx.Conversations.UpdateMessage(message)
+	if err != nil {
+		return nil, fmt.Errorf("Could not update message SelectedReply: %v", err)
+	}
+	return nextReply, nil
+}
+
+func (a *AppModel) PersistMessages() ([]conversation.Message, error) {
+	messages := make([]conversation.Message, len(a.Messages))
+	for i, m := range a.Messages {
+		if i == 0 && m.ID == 0 {
+			m.Conversation = &a.Conversation
+			m, err := a.Ctx.Conversations.SaveMessage(m)
+			if err != nil {
+				return nil, fmt.Errorf("Could not create first message %d: %v", a.Messages[i].ID, err)
+			}
+			messages[i] = *m
+			// let's set the conversation root message(s), as this is the first message
+			m.Conversation.RootMessages = []conversation.Message{*m}
+			m.Conversation.SelectedRoot = &m.Conversation.RootMessages[0]
+			a.Ctx.Conversations.UpdateConversation(m.Conversation)
+		} else if m.ID > 0 {
+			// Existing message, update it
+			err := a.Ctx.Conversations.UpdateMessage(&m)
+			if err != nil {
+				return nil, fmt.Errorf("Could not update message %d: %v", a.Messages[i].ID, err)
+			}
+			messages[i] = m
+		} else if i > 0 {
+			// New message, reply to previous
+			replies, err := a.Ctx.Conversations.Reply(&messages[i-1], m)
+			if err != nil {
+				return nil, fmt.Errorf("Could not reply with new message: %v", err)
+			}
+			messages[i] = replies[0]
+		} else {
+			return nil, fmt.Errorf("No messages to reply to (this is a bug)")
+		}
+	}
+	return messages, nil
+}
+
+func (a *AppModel) PersistConversation() (conversation.Conversation, error) {
+	conv := a.Conversation
+	var err error
+	if a.Conversation.ID > 0 {
+		err = a.Ctx.Conversations.UpdateConversation(&conv)
+	} else {
+		c, e := a.Ctx.Conversations.CreateConversation("")
+		err = e
+		if e == nil && c != nil {
+		  conv = *c
+		}
+	}
+	return conv, err
+}
+
+func (a *AppModel) ExecuteToolCalls(toolCalls []api.ToolCall) ([]api.ToolResult, error) {
+	agent := a.Ctx.GetAgent(a.Ctx.Config.Defaults.Agent)
+	if agent == nil {
+		return nil, fmt.Errorf("Attempted to execute tool calls with no agent configured")
+	}
+
+	return agents.ExecuteToolCalls(toolCalls, agent.Toolbox)
+}
+
+func (a *AppModel) Prompt(
+	messages []conversation.Message,
+	chatReplyChunks chan provider.Chunk,
+	stopSignal chan struct{},
+) (*conversation.Message, error) {
+	model, _, p, err := a.Ctx.GetModelProvider(a.Model, a.ProviderName)
+	if err != nil {
+		return nil, err
+	}
+
+	params := provider.RequestParameters{
+		Model:       model,
+		MaxTokens:   *a.Ctx.Config.Defaults.MaxTokens,
+		Temperature: *a.Ctx.Config.Defaults.Temperature,
+	}
+
+	if a.Agent != nil {
+		params.Toolbox = a.Agent.Toolbox
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+
+	go func() {
+		select {
+		case <-stopSignal:
+			cancel()
+		}
+	}()
+
+	msg, err := p.CreateChatCompletionStream(
+		ctx, params, conversation.MessagesToAPI(messages), chatReplyChunks,
+	)
+
+	if msg != nil {
+		msg := conversation.MessageFromAPI(*msg)
+		msg.Metadata.GenerationProvider = &a.ProviderName
+		msg.Metadata.GenerationModel = &a.Model
+		return &msg, err
+	}
+	return nil, err
+}
--- a/pkg/tui/shared/shared.go
+++ b/pkg/tui/shared/shared.go
@ -0,0 +1,66 @@
+package shared
+
+import (
+	tea "github.com/charmbracelet/bubbletea"
+)
+
+// An analogue to tea.Model with support for checking if the model has been
+// initialized before
+type ViewModel interface {
+	Init() tea.Cmd
+	Update(tea.Msg) (ViewModel, tea.Cmd)
+
+	// View methods
+	Header(width int) string
+	// Render the view's main content into a container of the given dimensions
+	Content(width, height int) string
+	Footer(width int) string
+}
+
+type View int
+
+const (
+	ViewChat View = iota
+	ViewConversations
+	ViewSettings
+	//StateHelp
+)
+
+type (
+	// send to change the current state
+	MsgViewChange View
+	// sent to a state when it is entered, with the view we're leaving
+	MsgViewEnter View
+	// sent when a recoverable error occurs (displayed to user)
+	MsgError struct { Err error }
+	// sent when the view has handled a key input
+	MsgKeyHandled tea.KeyMsg
+)
+
+func ViewEnter(from View) tea.Cmd {
+	return func() tea.Msg {
+		return MsgViewEnter(from)
+	}
+}
+
+func ChangeView(view View) tea.Cmd {
+	return func() tea.Msg {
+		return MsgViewChange(view)
+	}
+}
+
+func KeyHandled(key tea.KeyMsg) tea.Cmd {
+	return func() tea.Msg {
+		return MsgKeyHandled(key)
+	}
+}
+
+func WrapError(err error) tea.Cmd {
+	return func() tea.Msg {
+		return MsgError{ Err: err }
+	}
+}
+
+func AsMsgError(err error) MsgError {
+	return MsgError{ Err: err }
+}
--- a/pkg/tui/styles/styles.go
+++ b/pkg/tui/styles/styles.go
@ -0,0 +1,8 @@
+package styles
+
+import "github.com/charmbracelet/lipgloss"
+
+var Header = lipgloss.NewStyle().
+	PaddingLeft(1).
+	PaddingRight(1).
+	Background(lipgloss.Color("0"))
--- a/pkg/tui/tui.go
+++ b/pkg/tui/tui.go
@ -1,845 +1,163 @@
 package tui

-// The terminal UI for lmcli, launched from the `lmcli chat` command
-// TODO:
-// - conversation list view
-// - change model
-// - rename conversation
-// - set system prompt
-// - system prompt library?
-
 import (
-	"context"
 	"fmt"
-	"strings"
-	"time"

-	cmdutil "git.mlow.ca/mlow/lmcli/pkg/cmd/util"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
 	"git.mlow.ca/mlow/lmcli/pkg/lmcli"
-	models "git.mlow.ca/mlow/lmcli/pkg/lmcli/model"
-	"github.com/charmbracelet/bubbles/spinner"
-	"github.com/charmbracelet/bubbles/textarea"
-	"github.com/charmbracelet/bubbles/viewport"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/model"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	tuiutil "git.mlow.ca/mlow/lmcli/pkg/tui/util"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/views/chat"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/views/conversations"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/views/settings"
 	tea "github.com/charmbracelet/bubbletea"
 	"github.com/charmbracelet/lipgloss"
-	"github.com/muesli/reflow/wordwrap"
 )

-type focusState int
+type Model struct {
+	App *model.AppModel

-const (
-	focusInput focusState = iota
-	focusMessages
-)
-
-type editorTarget int
-
-const (
-	input editorTarget = iota
-	selectedMessage
-)
-
-type model struct {
+	// window size
 	width  int
 	height int

-	ctx           *lmcli.Context
-	convShortname string
+	// errors to display
+	// TODO: allow dismissing errors
+	errs []error

-	// application state
-	conversation    *models.Conversation
-	messages        []models.Message
-	waitingForReply bool
-	editorTarget    editorTarget
-	stopSignal      chan interface{}
-	replyChan       chan models.Message
-	replyChunkChan  chan string
-	persistence     bool // whether we will save new messages in the conversation
-	err             error
-
-	// ui state
-	focus           focusState
-	wrap            bool     // whether message content is wrapped to viewport width
-	status          string   // a general status message
-	highlightCache  []string // a cache of syntax highlighted message content
-	messageOffsets  []int
-	selectedMessage int
-
-	// ui elements
-	content viewport.Model
-	input   textarea.Model
-	spinner spinner.Model
+	activeView shared.View
+	views      map[shared.View]shared.ViewModel
 }

-type message struct {
-	role    string
-	content string
-}
+func initialModel(ctx *lmcli.Context, opts LaunchOptions) *Model {
+	app := model.NewAppModel(ctx, opts.InitialConversation)

-// custom tea.Msg types
-type (
-	// sent on each chunk received from LLM
-	msgResponseChunk string
-	// sent when response is finished being received
-	msgResponseEnd string
-	// a special case of msgError that stops the response waiting animation
-	msgResponseError error
-	// sent on each completed reply
-	msgAssistantReply models.Message
-	// sent when a conversation is (re)loaded
-	msgConversationLoaded *models.Conversation
-	// sent when a new conversation title is set
-	msgConversationTitleChanged string
-	// send when a conversation's messages are laoded
-	msgMessagesLoaded []models.Message
-	// sent when an error occurs
-	msgError error
-)
-
-// styles
-var (
-	userStyle      = lipgloss.NewStyle().Faint(true).Bold(true).Foreground(lipgloss.Color("10"))
-	assistantStyle = lipgloss.NewStyle().Faint(true).Bold(true).Foreground(lipgloss.Color("12"))
-	messageStyle   = lipgloss.NewStyle().PaddingLeft(2).PaddingRight(2)
-	headerStyle    = lipgloss.NewStyle().
-			Background(lipgloss.Color("0"))
-	conversationStyle = lipgloss.NewStyle().
-				MarginTop(1).
-				MarginBottom(1)
-	footerStyle = lipgloss.NewStyle().
-			BorderTop(true).
-			BorderStyle(lipgloss.NormalBorder())
-)
-
-func (m model) Init() tea.Cmd {
-	return tea.Batch(
-		textarea.Blink,
-		m.spinner.Tick,
-		m.loadConversation(m.convShortname),
-		m.waitForChunk(),
-		m.waitForReply(),
-	)
-}
-
-func wrapError(err error) tea.Cmd {
-	return func() tea.Msg {
-		return msgError(err)
-	}
-}
-
-func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	var cmds []tea.Cmd
-
-	switch msg := msg.(type) {
-	case msgTempfileEditorClosed:
-		contents := string(msg)
-		switch m.editorTarget {
-		case input:
-			m.input.SetValue(contents)
-		case selectedMessage:
-			m.setMessageContents(m.selectedMessage, contents)
-			if m.persistence && m.messages[m.selectedMessage].ID > 0 {
-				// update persisted message
-				err := m.ctx.Store.UpdateMessage(&m.messages[m.selectedMessage])
-				if err != nil {
-					cmds = append(cmds, wrapError(fmt.Errorf("Could not save edited message: %v", err)))
-				}
-			}
-			m.updateContent()
-		}
-	case tea.KeyMsg:
-		switch msg.String() {
-		case "ctrl+c":
-			if m.waitingForReply {
-				m.stopSignal <- ""
-			} else {
-				return m, tea.Quit
-			}
-		case "ctrl+p":
-			m.persistence = !m.persistence
-		case "ctrl+w":
-			m.wrap = !m.wrap
-			m.updateContent()
-		case "q":
-			if m.focus != focusInput {
-				return m, tea.Quit
-			}
-		default:
-			var inputHandled tea.Cmd
-			switch m.focus {
-			case focusInput:
-				inputHandled = m.handleInputKey(msg)
-			case focusMessages:
-				inputHandled = m.handleMessagesKey(msg)
-			}
-			if inputHandled != nil {
-				return m, inputHandled
-			}
-		}
-	case tea.WindowSizeMsg:
-		m.width = msg.Width
-		m.height = msg.Height
-		m.content.Width = msg.Width
-		m.content.Height = msg.Height - m.getFixedComponentHeight()
-		m.input.SetWidth(msg.Width - 1)
-		m.updateContent()
-	case msgConversationLoaded:
-		m.conversation = (*models.Conversation)(msg)
-		cmds = append(cmds, m.loadMessages(m.conversation))
-	case msgMessagesLoaded:
-		m.setMessages(msg)
-		m.updateContent()
-	case msgResponseChunk:
-		chunk := string(msg)
-		last := len(m.messages) - 1
-		if last >= 0 && m.messages[last].Role == models.MessageRoleAssistant {
-			m.setMessageContents(last, m.messages[last].Content+chunk)
-		} else {
-			m.addMessage(models.Message{
-				Role:    models.MessageRoleAssistant,
-				Content: chunk,
-			})
-		}
-		m.updateContent()
-		cmds = append(cmds, m.waitForChunk()) // wait for the next chunk
-	case msgAssistantReply:
-		// the last reply that was being worked on is finished
-		reply := models.Message(msg)
-		last := len(m.messages) - 1
-		if last < 0 {
-			panic("Unexpected empty messages handling msgReply")
-		}
-		m.setMessageContents(last, strings.TrimSpace(m.messages[last].Content))
-		if m.messages[last].Role == models.MessageRoleAssistant {
-			// the last message was an assistant message, so this is a continuation
-			if reply.Role == models.MessageRoleToolCall {
-				// update last message rrole to tool call
-				m.messages[last].Role = models.MessageRoleToolCall
-			}
-		} else {
-			m.addMessage(reply)
-		}
-
-		if m.persistence {
-			var err error
-			if m.conversation.ID == 0 {
-				err = m.ctx.Store.SaveConversation(m.conversation)
-			}
-			if err != nil {
-				cmds = append(cmds, wrapError(err))
-			} else {
-				cmds = append(cmds, m.persistConversation())
-			}
-		}
-
-		if m.conversation.Title == "" {
-			cmds = append(cmds, m.generateConversationTitle())
-		}
-
-		m.updateContent()
-		cmds = append(cmds, m.waitForReply())
-	case msgResponseEnd:
-		m.waitingForReply = false
-		last := len(m.messages) - 1
-		if last < 0 {
-			panic("Unexpected empty messages handling msgResponseEnd")
-		}
-		m.setMessageContents(last, strings.TrimSpace(m.messages[last].Content))
-		m.updateContent()
-		m.status = "Press ctrl+s to send"
-	case msgResponseError:
-		m.waitingForReply = false
-		m.status = "Press ctrl+s to send"
-		m.err = error(msg)
-	case msgConversationTitleChanged:
-		title := string(msg)
-		m.conversation.Title = title
-		if m.persistence {
-			err := m.ctx.Store.SaveConversation(m.conversation)
-			if err != nil {
-				cmds = append(cmds, wrapError(err))
-			}
-		}
-	case msgError:
-		m.err = error(msg)
-	}
-
-	var cmd tea.Cmd
-	m.spinner, cmd = m.spinner.Update(msg)
-	if cmd != nil {
-		cmds = append(cmds, cmd)
-	}
-
-	inputCaptured := false
-	m.input, cmd = m.input.Update(msg)
-	if cmd != nil {
-		inputCaptured = true
-		cmds = append(cmds, cmd)
-	}
-
-	if !inputCaptured {
-		m.content, cmd = m.content.Update(msg)
-		if cmd != nil {
-			cmds = append(cmds, cmd)
-		}
-	}
-
-	return m, tea.Batch(cmds...)
-}
-
-func (m model) View() string {
-	if m.width == 0 {
-		// this is the case upon initial startup, but it's also a safe bet that
-		// we can just skip rendering if the terminal is really 0 width...
-		// without this, the m.*View() functions may crash
-		return ""
-	}
-
-	sections := make([]string, 0, 6)
-	sections = append(sections, m.headerView())
-	sections = append(sections, m.contentView())
-	error := m.errorView()
-	if error != "" {
-		sections = append(sections, error)
-	}
-	sections = append(sections, m.inputView())
-	sections = append(sections, m.footerView())
-
-	return lipgloss.JoinVertical(
-		lipgloss.Left,
-		sections...,
-	)
-}
-
-// returns the total height of "fixed" components, which are those which don't
-// change height dependent on window size.
-func (m *model) getFixedComponentHeight() int {
-	h := 0
-	h += m.input.Height()
-	h += lipgloss.Height(m.headerView())
-	h += lipgloss.Height(m.footerView())
-	errorView := m.errorView()
-	if errorView != "" {
-		h += lipgloss.Height(errorView)
-	}
-	return h
-}
-
-func (m *model) headerView() string {
-	titleStyle := lipgloss.NewStyle().
-		PaddingLeft(1).
-		PaddingRight(1).
-		Bold(true)
-	var title string
-	if m.conversation != nil && m.conversation.Title != "" {
-		title = m.conversation.Title
-	} else {
-		title = "Untitled"
-	}
-	part := titleStyle.Render(title)
-
-	return headerStyle.Width(m.width).Render(part)
-}
-
-func (m *model) contentView() string {
-	return m.content.View()
-}
-
-func (m *model) errorView() string {
-	if m.err == nil {
-		return ""
-	}
-	return lipgloss.NewStyle().
-		Width(m.width).
-		AlignHorizontal(lipgloss.Center).
-		Bold(true).
-		Foreground(lipgloss.Color("1")).
-		Render(fmt.Sprintf("%s", m.err))
-}
-
-func (m *model) inputView() string {
-	return m.input.View()
-}
-
-func (m *model) footerView() string {
-	segmentStyle := lipgloss.NewStyle().PaddingLeft(1).PaddingRight(1).Faint(true)
-	segmentSeparator := "|"
-
-	savingStyle := segmentStyle.Copy().Bold(true)
-	saving := ""
-	if m.persistence {
-		saving = savingStyle.Foreground(lipgloss.Color("2")).Render("✅💾")
-	} else {
-		saving = savingStyle.Foreground(lipgloss.Color("1")).Render("❌💾")
-	}
-
-	status := m.status
-	if m.waitingForReply {
-		status += m.spinner.View()
-	}
-
-	leftSegments := []string{
-		saving,
-		segmentStyle.Render(status),
-	}
-	rightSegments := []string{
-		segmentStyle.Render(fmt.Sprintf("Model: %s", *m.ctx.Config.Defaults.Model)),
-	}
-
-	left := strings.Join(leftSegments, segmentSeparator)
-	right := strings.Join(rightSegments, segmentSeparator)
-
-	totalWidth := lipgloss.Width(left) + lipgloss.Width(right)
-	remaining := m.width - totalWidth
-
-	var padding string
-	if remaining > 0 {
-		padding = strings.Repeat(" ", remaining)
-	}
-
-	footer := left + padding + right
-	if remaining < 0 {
-		ellipses := "... "
-		// this doesn't work very well, due to trying to trim a string with
-		// ansii chars already in it
-		footer = footer[:(len(footer)+remaining)-len(ellipses)-3] + ellipses
-	}
-	return footerStyle.Width(m.width).Render(footer)
-}
-
-func initialModel(ctx *lmcli.Context, convShortname string) model {
-	m := model{
-		ctx:           ctx,
-		convShortname: convShortname,
-		conversation:  &models.Conversation{},
-		persistence:   true,
-
-		stopSignal:     make(chan interface{}),
-		replyChan:      make(chan models.Message),
-		replyChunkChan: make(chan string),
-
-		wrap:            true,
-		selectedMessage: -1,
-	}
-
-	m.content = viewport.New(0, 0)
-
-	m.input = textarea.New()
-	m.input.CharLimit = 0
-	m.input.Placeholder = "Enter a message"
-
-	m.input.FocusedStyle.CursorLine = lipgloss.NewStyle()
-	m.input.ShowLineNumbers = false
-	m.input.SetHeight(4)
-	m.input.Focus()
-
-	m.spinner = spinner.New(spinner.WithSpinner(
-		spinner.Spinner{
-			Frames: []string{
-				".  ",
-				".. ",
-				"...",
-				".. ",
-				".  ",
-				"   ",
-			},
-			FPS: time.Second / 3,
+	m := Model{
+		App:        app,
+		activeView: opts.InitialView,
+		views: map[shared.View]shared.ViewModel{
+			shared.ViewChat:          chat.Chat(app),
+			shared.ViewConversations: conversations.Conversations(app),
+			shared.ViewSettings:      settings.Settings(app),
 		},
-	))
+	}

-	m.waitingForReply = false
-	m.status = "Press ctrl+s to send"
-	return m
+	return &m
 }

-// fraction is the fraction of the total screen height into view the offset
-// should be scrolled into view. 0.5 = items will be snapped to middle of
-// view
-func scrollIntoView(vp *viewport.Model, offset int, fraction float32) {
-	currentOffset := vp.YOffset
-	if offset >= currentOffset && offset < currentOffset+vp.Height {
-		return
-	}
-	distance := currentOffset - offset
-	if distance < 0 {
-		// we should scroll down until it just comes into view
-		vp.SetYOffset(currentOffset - (distance + (vp.Height - int(float32(vp.Height)*fraction))) + 1)
-	} else {
-		// we should scroll up
-		vp.SetYOffset(currentOffset - distance - int(float32(vp.Height)*fraction))
+func (m *Model) Init() tea.Cmd {
+	var cmds []tea.Cmd
+	for _, v := range m.views {
+		// Init views
+		cmds = append(cmds, v.Init())
 	}
+	cmds = append(cmds, func() tea.Msg {
+		// Initial view change
+		return shared.MsgViewChange(m.activeView)
+	})
+	return tea.Batch(cmds...)
 }

-func (m *model) handleMessagesKey(msg tea.KeyMsg) tea.Cmd {
-	switch msg.String() {
-	case "tab":
-		m.focus = focusInput
-		m.updateContent()
-		m.input.Focus()
-	case "e":
-		message := m.messages[m.selectedMessage]
-		cmd := openTempfileEditor("message.*.md", message.Content, "# Edit the message below\n")
-		m.editorTarget = selectedMessage
-		return cmd
-	case "ctrl+k":
-		if m.selectedMessage > 0 && len(m.messages) == len(m.messageOffsets) {
-			m.selectedMessage--
-			m.updateContent()
-			offset := m.messageOffsets[m.selectedMessage]
-			scrollIntoView(&m.content, offset, 0.1)
-		}
-	case "ctrl+j":
-		if m.selectedMessage < len(m.messages)-1 && len(m.messages) == len(m.messageOffsets) {
-			m.selectedMessage++
-			m.updateContent()
-			offset := m.messageOffsets[m.selectedMessage]
-			scrollIntoView(&m.content, offset, 0.1)
-		}
-	case "ctrl+r":
-		// resubmit the conversation with all messages up until and including
-		// the selected message
-		if len(m.messages) == 0 {
-			return nil
-		}
-		m.messages = m.messages[:m.selectedMessage+1]
-		m.highlightCache = m.highlightCache[:m.selectedMessage+1]
-		m.updateContent()
-		m.content.GotoBottom()
-		return m.promptLLM()
-	}
-	return nil
-}
-
-func (m *model) handleInputKey(msg tea.KeyMsg) tea.Cmd {
-	switch msg.String() {
-	case "esc":
-		m.focus = focusMessages
-		if m.selectedMessage < 0 || m.selectedMessage >= len(m.messages) {
-			m.selectedMessage = len(m.messages) - 1
-		}
-		m.updateContent()
-		m.input.Blur()
-	case "ctrl+s":
-		userInput := strings.TrimSpace(m.input.Value())
-		if strings.TrimSpace(userInput) == "" {
-			return nil
-		}
-
-		if len(m.messages) > 0 && m.messages[len(m.messages)-1].Role == models.MessageRoleUser {
-			return wrapError(fmt.Errorf("Can't reply to a user message"))
-		}
-
-		reply := models.Message{
-			Role:    models.MessageRoleUser,
-			Content: userInput,
-		}
-
-		if m.persistence {
-			var err error
-			if m.conversation.ID == 0 {
-				err = m.ctx.Store.SaveConversation(m.conversation)
-			}
-			if err != nil {
-				return wrapError(err)
-			}
-
-			// ensure all messages up to the one we're about to add are
-			// persistent
-			cmd := m.persistConversation()
-			if cmd != nil {
-				return cmd
-			}
-			// persist our new message, returning with any possible errors
-			savedReply, err := m.ctx.Store.AddReply(m.conversation, reply)
-			if err != nil {
-				return wrapError(err)
-			}
-			reply = *savedReply
-		}
-
-		m.input.SetValue("")
-		m.addMessage(reply)
-
-		m.updateContent()
-		m.content.GotoBottom()
-		return m.promptLLM()
-	case "ctrl+e":
-		cmd := openTempfileEditor("message.*.md", m.input.Value(), "# Edit your input below\n")
-		m.editorTarget = input
+func (m *Model) handleGlobalInput(msg tea.KeyMsg) tea.Cmd {
+	view, cmd := m.views[m.activeView].Update(msg)
+	m.views[m.activeView] = view
+	if cmd != nil {
 		return cmd
 	}
-	return nil
-}

-func (m *model) loadConversation(shortname string) tea.Cmd {
-	return func() tea.Msg {
-		if shortname == "" {
-			return nil
-		}
-		c, err := m.ctx.Store.ConversationByShortName(shortname)
-		if err != nil {
-			return msgError(fmt.Errorf("Could not lookup conversation: %v", err))
-		}
-		if c.ID == 0 {
-			return msgError(fmt.Errorf("Conversation not found: %s", shortname))
-		}
-		return msgConversationLoaded(c)
-	}
-}
-
-func (m *model) loadMessages(c *models.Conversation) tea.Cmd {
-	return func() tea.Msg {
-		messages, err := m.ctx.Store.Messages(c)
-		if err != nil {
-			return msgError(fmt.Errorf("Could not load conversation messages: %v\n", err))
-		}
-		return msgMessagesLoaded(messages)
-	}
-}
-
-func (m *model) waitForReply() tea.Cmd {
-	return func() tea.Msg {
-		return msgAssistantReply(<-m.replyChan)
-	}
-}
-
-func (m *model) waitForChunk() tea.Cmd {
-	return func() tea.Msg {
-		return msgResponseChunk(<-m.replyChunkChan)
-	}
-}
-
-func (m *model) generateConversationTitle() tea.Cmd {
-	return func() tea.Msg {
-		title, err := cmdutil.GenerateTitle(m.ctx, m.conversation)
-		if err != nil {
-			return msgError(err)
-		}
-		return msgConversationTitleChanged(title)
-	}
-}
-
-func (m *model) promptLLM() tea.Cmd {
-	m.waitingForReply = true
-	m.status = "Press ctrl+c to cancel"
-
-	return func() tea.Msg {
-		completionProvider, err := m.ctx.GetCompletionProvider(*m.ctx.Config.Defaults.Model)
-		if err != nil {
-			return msgError(err)
-		}
-
-		requestParams := models.RequestParameters{
-			Model:       *m.ctx.Config.Defaults.Model,
-			MaxTokens:   *m.ctx.Config.Defaults.MaxTokens,
-			Temperature: *m.ctx.Config.Defaults.Temperature,
-			ToolBag:     m.ctx.EnabledTools,
-		}
-
-		replyHandler := func(msg models.Message) {
-			m.replyChan <- msg
-		}
-
-		ctx, cancel := context.WithCancel(context.Background())
-
-		canceled := false
-		go func() {
-			select {
-			case <-m.stopSignal:
-				canceled = true
-				cancel()
-			}
-		}()
-
-		resp, err := completionProvider.CreateChatCompletionStream(
-			ctx, requestParams, m.messages, replyHandler, m.replyChunkChan,
-		)
-
-		if err != nil && !canceled {
-			return msgResponseError(err)
-		}
-
-		return msgResponseEnd(resp)
-	}
-}
-
-func (m *model) persistConversation() tea.Cmd {
-	existingMessages, err := m.ctx.Store.Messages(m.conversation)
-	if err != nil {
-		return wrapError(fmt.Errorf("Could not retrieve existing conversation messages while trying to save: %v", err))
-	}
-
-	existingById := make(map[uint]*models.Message, len(existingMessages))
-	for _, msg := range existingMessages {
-		existingById[msg.ID] = &msg
-	}
-
-	currentById := make(map[uint]*models.Message, len(m.messages))
-	for _, msg := range m.messages {
-		currentById[msg.ID] = &msg
-	}
-
-	for _, msg := range existingMessages {
-		_, ok := currentById[msg.ID]
-		if !ok {
-			err := m.ctx.Store.DeleteMessage(&msg)
-			if err != nil {
-				return wrapError(fmt.Errorf("Failed to remove messages: %v", err))
-			}
-		}
-	}
-
-	for i, msg := range m.messages {
-		if msg.ID > 0 {
-			exist, ok := existingById[msg.ID]
-			if ok {
-				if msg.Content == exist.Content {
-					continue
-				}
-				// update message when contents don't match that of store
-				err := m.ctx.Store.UpdateMessage(&msg)
-				if err != nil {
-					return wrapError(err)
-				}
-			} else {
-				// this would be quite odd... and I'm not sure how to handle
-				// it at the time of writing this
-			}
-		} else {
-			newMessage, err := m.ctx.Store.AddReply(m.conversation, msg)
-			if err != nil {
-				return wrapError(err)
-			}
-			m.setMessage(i, *newMessage)
-		}
+	switch msg.String() {
+	case "ctrl+c", "ctrl+q":
+		return tea.Quit
 	}
 	return nil
 }

-func (m *model) setMessages(messages []models.Message) {
-	m.messages = messages
-	m.highlightCache = make([]string, len(messages))
-	for i, msg := range m.messages {
-		highlighted, _ := m.ctx.Chroma.HighlightS(msg.Content)
-		m.highlightCache[i] = highlighted
+func (m *Model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
+	switch msg := msg.(type) {
+	case tea.WindowSizeMsg:
+		m.width, m.height = msg.Width, msg.Height
+	case tea.KeyMsg:
+		cmd := m.handleGlobalInput(msg)
+		if cmd != nil {
+			return m, cmd
+		}
+	case shared.MsgViewChange:
+		currView := m.activeView
+		m.activeView = shared.View(msg)
+		return m, tea.Batch(tea.WindowSize(), shared.ViewEnter(currView))
+	case shared.MsgError:
+		m.errs = append(m.errs, msg.Err)
+	}
+
+	view, cmd := m.views[m.activeView].Update(msg)
+	m.views[m.activeView] = view
+	return m, cmd
+}
+
+func (m *Model) View() string {
+	if m.width == 0 || m.height == 0 {
+		// we're dimensionless!
+		return ""
+	}
+
+	header := m.views[m.activeView].Header(m.width)
+	footer := m.views[m.activeView].Footer(m.width)
+	fixedUIHeight := tuiutil.Height(header) + tuiutil.Height(footer)
+
+	errBanners := make([]string, len(m.errs))
+	for idx, err := range m.errs {
+		errBanners[idx] = tuiutil.ErrorBanner(err, m.width)
+		fixedUIHeight += tuiutil.Height(errBanners[idx])
+	}
+
+	content := m.views[m.activeView].Content(m.width, m.height-fixedUIHeight)
+
+	sections := make([]string, 0, 4)
+	if header != "" {
+		sections = append(sections, header)
+	}
+	if content != "" {
+		sections = append(sections, content)
+	}
+	if footer != "" {
+		sections = append(sections, footer)
+	}
+	for _, errBanner := range errBanners {
+		sections = append(sections, errBanner)
+	}
+	return lipgloss.JoinVertical(lipgloss.Left, sections...)
+}
+
+type LaunchOptions struct {
+	InitialConversation *conversation.Conversation
+	InitialView         shared.View
+}
+
+type LaunchOption func(*LaunchOptions)
+
+func WithInitialConversation(conv *conversation.Conversation) LaunchOption {
+	return func(opts *LaunchOptions) {
+		opts.InitialConversation = conv
 	}
 }

-func (m *model) setMessage(i int, msg models.Message) {
-	if i >= len(m.messages) {
-		panic("i out of range")
-	}
-	highlighted, _ := m.ctx.Chroma.HighlightS(msg.Content)
-	m.messages[i] = msg
-	m.highlightCache[i] = highlighted
-}
-
-func (m *model) addMessage(msg models.Message) {
-	highlighted, _ := m.ctx.Chroma.HighlightS(msg.Content)
-	m.messages = append(m.messages, msg)
-	m.highlightCache = append(m.highlightCache, highlighted)
-}
-
-func (m *model) setMessageContents(i int, content string) {
-	if i >= len(m.messages) {
-		panic("i out of range")
-	}
-	highlighted, _ := m.ctx.Chroma.HighlightS(content)
-	m.messages[i].Content = content
-	m.highlightCache[i] = highlighted
-}
-
-func (m *model) updateContent() {
-	atBottom := m.content.AtBottom()
-	m.content.SetContent(m.conversationView())
-	if atBottom {
-		// if we were at bottom before the update, scroll with the output
-		m.content.GotoBottom()
+func WithInitialView(view shared.View) LaunchOption {
+	return func(opts *LaunchOptions) {
+		opts.InitialView = view
 	}
 }

-// render the conversation into a string
-func (m *model) conversationView() string {
-	sb := strings.Builder{}
-	msgCnt := len(m.messages)
-
-	m.messageOffsets = make([]int, len(m.messages))
-	lineCnt := conversationStyle.GetMarginTop()
-	for i, message := range m.messages {
-		m.messageOffsets[i] = lineCnt
-
-		icon := "⚙️"
-		friendly := message.Role.FriendlyRole()
-		style := lipgloss.NewStyle().Bold(true).Faint(true)
-
-		switch message.Role {
-		case models.MessageRoleUser:
-			icon = ""
-			style = userStyle
-		case models.MessageRoleAssistant:
-			icon = ""
-			style = assistantStyle
-		case models.MessageRoleToolCall, models.MessageRoleToolResult:
-			icon = "🔧"
-		}
-
-		// write message heading with space for content
-		user := style.Render(icon + friendly)
-
-		var prefix string
-		var suffix string
-
-		faint := lipgloss.NewStyle().Faint(true)
-		if m.focus == focusMessages {
-			if i == m.selectedMessage {
-				prefix = "> "
-			}
-			suffix += faint.Render(fmt.Sprintf(" (%d/%d)", i+1, msgCnt))
-		}
-
-		if message.ID == 0 {
-			suffix += faint.Render(" (not saved)")
-		}
-
-		header := lipgloss.NewStyle().PaddingLeft(1).Render(prefix + user + suffix)
-		sb.WriteString(header)
-		lineCnt += lipgloss.Height(header)
-
-		// TODO: special rendering for tool calls/results?
-		if message.Content != "" {
-			sb.WriteString("\n\n")
-			lineCnt += 1
-
-			// write message contents
-			var highlighted string
-			if m.highlightCache[i] == "" {
-				highlighted = message.Content
-			} else {
-				highlighted = m.highlightCache[i]
-			}
-			var contents string
-			if m.wrap {
-				wrapWidth := m.content.Width - messageStyle.GetHorizontalPadding() - 2
-				wrapped := wordwrap.String(highlighted, wrapWidth)
-				contents = wrapped
-			} else {
-				contents = highlighted
-			}
-			sb.WriteString(messageStyle.Width(0).Render(contents))
-			lineCnt += lipgloss.Height(contents)
-		}
-
-		if i < msgCnt-1 {
-			sb.WriteString("\n\n")
-			lineCnt += 1
-		}
+func Launch(ctx *lmcli.Context, options ...LaunchOption) error {
+	opts := &LaunchOptions{
+		InitialView: shared.ViewChat,
+	}
+	for _, opt := range options {
+		opt(opts)
 	}
-	return conversationStyle.Render(sb.String())
-}

-func Launch(ctx *lmcli.Context, convShortname string) error {
-	p := tea.NewProgram(initialModel(ctx, convShortname), tea.WithAltScreen())
-	if _, err := p.Run(); err != nil {
+	program := tea.NewProgram(initialModel(ctx, *opts), tea.WithAltScreen())
+	if _, err := program.Run(); err != nil {
 		return fmt.Errorf("Error running program: %v", err)
 	}
 	return nil
--- a/pkg/tui/util.go
+++ b/pkg/tui/util.go
@ -1,42 +0,0 @@
-package tui
-
-import (
-	"os"
-	"os/exec"
-	"strings"
-
-	tea "github.com/charmbracelet/bubbletea"
-)
-
-type msgTempfileEditorClosed string
-
-// openTempfileEditor opens an $EDITOR on a new temporary file with the given
-// content. Upon closing, the contents of the file are read back returned
-// wrapped in a msgTempfileEditorClosed returned by the tea.Cmd
-func openTempfileEditor(pattern string, content string, placeholder string) tea.Cmd {
-	msgFile, _ := os.CreateTemp("/tmp", pattern)
-
-	err := os.WriteFile(msgFile.Name(), []byte(placeholder+content), os.ModeAppend)
-	if err != nil {
-		return wrapError(err)
-	}
-
-	editor := os.Getenv("EDITOR")
-	if editor == "" {
-		editor = "vim"
-	}
-
-	c := exec.Command(editor, msgFile.Name())
-	return tea.ExecProcess(c, func(err error) tea.Msg {
-		bytes, err := os.ReadFile(msgFile.Name())
-		if err != nil {
-			return msgError(err)
-		}
-		fileContents := string(bytes)
-		if strings.HasPrefix(fileContents, placeholder) {
-			fileContents = fileContents[len(placeholder):]
-		}
-		stripped := strings.Trim(fileContents, "\n \t")
-		return msgTempfileEditorClosed(stripped)
-	})
-}
--- a/pkg/tui/util/util.go
+++ b/pkg/tui/util/util.go
@ -0,0 +1,137 @@
+package util
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+	"github.com/muesli/reflow/ansi"
+)
+
+type MsgTempfileEditorClosed string
+
+// OpenTempfileEditor opens $EDITOR on a temporary file with the given content.
+// Upon closing, the contents of the file are read and returned wrapped in a
+// MsgTempfileEditorClosed
+func OpenTempfileEditor(pattern string, content string, placeholder string) tea.Cmd {
+	msgFile, _ := os.CreateTemp("/tmp", pattern)
+
+	err := os.WriteFile(msgFile.Name(), []byte(placeholder+content), os.ModeAppend)
+	if err != nil {
+		return func() tea.Msg { return err }
+	}
+
+	editor := os.Getenv("EDITOR")
+	if editor == "" {
+		editor = "vim"
+	}
+
+	c := exec.Command(editor, msgFile.Name())
+	return tea.ExecProcess(c, func(err error) tea.Msg {
+		bytes, err := os.ReadFile(msgFile.Name())
+		if err != nil {
+			return err
+		}
+		os.Remove(msgFile.Name())
+		fileContents := string(bytes)
+		if strings.HasPrefix(fileContents, placeholder) {
+			fileContents = fileContents[len(placeholder):]
+		}
+		stripped := strings.Trim(fileContents, "\n \t")
+		return MsgTempfileEditorClosed(stripped)
+	})
+}
+
+// similar to lipgloss.Height, except returns 0 instead of 1 on empty strings
+func Height(str string) int {
+	if str == "" {
+		return 0
+	}
+	return strings.Count(str, "\n") + 1
+}
+
+func Width(str string) int {
+	if str == "" {
+		return 0
+	}
+	return ansi.PrintableRuneWidth(str)
+}
+
+func TruncateRightToCellWidth(str string, width int, tail string) string {
+    cellWidth := ansi.PrintableRuneWidth(str)
+    if cellWidth <= width {
+        return str
+    }
+
+    tailWidth := ansi.PrintableRuneWidth(tail)
+    if width <= tailWidth {
+        return tail[:width]
+    }
+
+    targetWidth := width - tailWidth
+    runes := []rune(str)
+
+    for i := len(runes) - 1; i >= 0; i-- {
+        str = string(runes[:i])
+        if ansi.PrintableRuneWidth(str) <= targetWidth {
+            return str + tail
+        }
+    }
+
+    return tail
+}
+
+func TruncateLeftToCellWidth(str string, width int, tail string) string {
+    cellWidth := ansi.PrintableRuneWidth(str)
+    if cellWidth <= width {
+        return str
+    }
+
+    tailWidth := ansi.PrintableRuneWidth(tail)
+    if width <= tailWidth {
+        return tail[:width]
+    }
+
+    targetWidth := width - tailWidth
+    runes := []rune(str)
+
+    for i := 0; i < len(runes); i++ {
+        str = string(runes[i:])
+        if ansi.PrintableRuneWidth(str) <= targetWidth {
+            return tail + str
+        }
+    }
+
+    return tail
+}
+
+func ScrollIntoView(vp *viewport.Model, offset int, edge int) {
+	currentOffset := vp.YOffset
+	if offset >= currentOffset && offset < currentOffset+vp.Height {
+		return
+	}
+	distance := currentOffset - offset
+	if distance < 0 {
+		// we should scroll down until it just comes into view
+		vp.SetYOffset(currentOffset - (distance + (vp.Height - edge)) + 1)
+	} else {
+		// we should scroll up
+		vp.SetYOffset(currentOffset - distance - edge)
+	}
+}
+
+func ErrorBanner(err error, width int) string {
+	if err == nil {
+		return ""
+	}
+	return lipgloss.NewStyle().
+		Width(width).
+		AlignHorizontal(lipgloss.Center).
+		Bold(true).
+		Foreground(lipgloss.Color("1")).
+		Render(fmt.Sprintf("%s", err))
+}
--- a/pkg/tui/views/chat/chat.go
+++ b/pkg/tui/views/chat/chat.go
@ -0,0 +1,168 @@
+package chat
+
+import (
+	"time"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/provider"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/model"
+	"github.com/charmbracelet/bubbles/cursor"
+	"github.com/charmbracelet/bubbles/spinner"
+	"github.com/charmbracelet/bubbles/textarea"
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+// custom tea.Msg types
+type (
+	// sent when a new conversation title generated
+	msgConversationTitleGenerated string
+	// sent when the conversation has been persisted, triggers a reload of contents
+	msgConversationPersisted conversation.Conversation
+	msgMessagesPersisted []conversation.Message
+	// sent when a conversation's messages are laoded
+	msgConversationMessagesLoaded struct {
+		messages     []conversation.Message
+	}
+	// a special case of common.MsgError that stops the response waiting animation
+	msgChatResponseError struct {
+		Err error
+	}
+	// sent on each chunk received from LLM
+	msgChatResponseChunk provider.Chunk
+	// sent on each completed reply
+	msgChatResponse conversation.Message
+	// sent when the response is canceled
+	msgChatResponseCanceled struct{}
+	// sent when results from a tool call are returned
+	msgToolResults []api.ToolResult
+	// sent when the given message is made the new selected reply of its parent
+	msgSelectedReplyCycled *conversation.Message
+	// sent when the given message is made the new selected root of the current conversation
+	msgSelectedRootCycled *conversation.Message
+	// sent when a message's contents are updated and saved
+	msgMessageUpdated *conversation.Message
+	// sent when a message is cloned, with the cloned message
+	msgMessageCloned *conversation.Message
+)
+
+type focusState int
+
+const (
+	focusInput focusState = iota
+	focusMessages
+)
+
+type editorTarget int
+
+const (
+	input editorTarget = iota
+	selectedMessage
+)
+
+type state int
+
+const (
+	idle state = iota
+	loading
+	pendingResponse
+)
+
+type Model struct {
+	// App state
+	App    *model.AppModel
+	Height int
+	Width  int
+
+	// Chat view state
+	state           state // current overall status of the view
+	selectedMessage int
+	editorTarget    editorTarget
+	stopSignal      chan struct{}
+	replyChan       chan conversation.Message
+	chatReplyChunks chan provider.Chunk
+	persistence     bool // whether we will save new messages in the conversation
+
+	// UI state
+	focus          focusState
+	showDetails    bool     // whether various details are shown in the UI (e.g. system prompt, tool calls/results, message metadata)
+	wrap           bool     // whether message content is wrapped to viewport width
+	messageCache   []string // cache of syntax highlighted and wrapped message content
+	messageOffsets []int
+
+	// ui elements
+	content     viewport.Model
+	input       textarea.Model
+	spinner     spinner.Model
+	replyCursor cursor.Model // cursor to indicate incoming response
+
+	// metrics
+	tokenCount uint
+	startTime  time.Time
+	elapsed    time.Duration
+}
+
+func getSpinner() spinner.Model {
+	return spinner.New(spinner.WithSpinner(
+		spinner.Spinner{
+			Frames: []string{
+				"∙∙∙",
+				"●∙∙",
+				"●●∙",
+				"●●●",
+				"∙●●",
+				"∙∙●",
+				"∙∙∙",
+				"∙∙●",
+				"∙●●",
+				"●●●",
+				"●●∙",
+				"●∙∙",
+			},
+			FPS: 440 * time.Millisecond,
+		},
+	))
+}
+
+func Chat(app *model.AppModel) *Model {
+	m := Model{
+		App: app,
+
+		state:       idle,
+		persistence: true,
+
+		stopSignal:      make(chan struct{}),
+		replyChan:       make(chan conversation.Message),
+		chatReplyChunks: make(chan provider.Chunk),
+
+		wrap:            true,
+		selectedMessage: -1,
+
+		content:     viewport.New(0, 0),
+		input:       textarea.New(),
+		spinner:     getSpinner(),
+		replyCursor: cursor.New(),
+	}
+
+	m.replyCursor.SetChar(" ")
+	m.replyCursor.Focus()
+
+	m.input.Focus()
+	m.input.MaxHeight = 0
+	m.input.CharLimit = 0
+	m.input.ShowLineNumbers = false
+	m.input.Placeholder = "Enter a message"
+
+	m.input.FocusedStyle.CursorLine = lipgloss.NewStyle()
+	m.input.FocusedStyle.Base = inputFocusedStyle
+	m.input.BlurredStyle.Base = inputBlurredStyle
+	return &m
+}
+
+func (m *Model) Init() tea.Cmd {
+	return tea.Batch(
+		m.waitForResponseChunk(),
+	)
+}
--- a/pkg/tui/views/chat/cmds.go
+++ b/pkg/tui/views/chat/cmds.go
@ -0,0 +1,136 @@
+package chat
+
+import (
+	"time"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/model"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	tea "github.com/charmbracelet/bubbletea"
+)
+
+func (m *Model) waitForResponseChunk() tea.Cmd {
+	return func() tea.Msg {
+		return msgChatResponseChunk(<-m.chatReplyChunks)
+	}
+}
+
+func (m *Model) loadConversationMessages() tea.Cmd {
+	return func() tea.Msg {
+		messages, err := m.App.LoadConversationMessages()
+		if err != nil {
+			return shared.AsMsgError(err)
+		}
+		return msgConversationMessagesLoaded{messages}
+	}
+}
+
+func (m *Model) generateConversationTitle() tea.Cmd {
+	return func() tea.Msg {
+		title, err := m.App.GenerateConversationTitle(m.App.Messages)
+		if err != nil {
+			return shared.AsMsgError(err)
+		}
+		return msgConversationTitleGenerated(title)
+	}
+}
+
+func (m *Model) cloneMessage(message conversation.Message, selected bool) tea.Cmd {
+	return func() tea.Msg {
+		msg, err := m.App.CloneMessage(message, selected)
+		if err != nil {
+			return shared.WrapError(err)
+		}
+		return msgMessageCloned(msg)
+	}
+}
+
+func (m *Model) updateMessageContent(message *conversation.Message) tea.Cmd {
+	return func() tea.Msg {
+		err := m.App.UpdateMessageContent(message)
+		if err != nil {
+			return shared.WrapError(err)
+		}
+		return msgMessageUpdated(message)
+	}
+}
+
+func (m *Model) cycleSelectedRoot(conv *conversation.Conversation, dir model.MessageCycleDirection) tea.Cmd {
+	if len(conv.RootMessages) < 2 {
+		return nil
+	}
+
+	return func() tea.Msg {
+		nextRoot, err := m.App.CycleSelectedRoot(conv, dir)
+		if err != nil {
+			return shared.WrapError(err)
+		}
+		return msgSelectedRootCycled(nextRoot)
+	}
+}
+
+func (m *Model) cycleSelectedReply(message *conversation.Message, dir model.MessageCycleDirection) tea.Cmd {
+	if len(message.Replies) < 2 {
+		return nil
+	}
+
+	return func() tea.Msg {
+		nextReply, err := m.App.CycleSelectedReply(message, dir)
+		if err != nil {
+			return shared.WrapError(err)
+		}
+		return msgSelectedReplyCycled(nextReply)
+	}
+}
+
+func (m *Model) persistConversation() tea.Cmd {
+	return func() tea.Msg {
+		conversation, err := m.App.PersistConversation()
+		if err != nil {
+			return shared.AsMsgError(err)
+		}
+		return msgConversationPersisted(conversation)
+	}
+}
+
+func (m *Model) persistMessages() tea.Cmd {
+	return func() tea.Msg {
+		messages, err := m.App.PersistMessages()
+		if err != nil {
+			return shared.AsMsgError(err)
+		}
+		return msgMessagesPersisted(messages)
+	}
+}
+
+func (m *Model) executeToolCalls(toolCalls []api.ToolCall) tea.Cmd {
+	return func() tea.Msg {
+		results, err := m.App.ExecuteToolCalls(toolCalls)
+		if err != nil {
+			return shared.AsMsgError(err)
+		}
+		return msgToolResults(results)
+	}
+}
+
+func (m *Model) promptLLM() tea.Cmd {
+	m.state = pendingResponse
+	m.spinner = getSpinner()
+	m.replyCursor.Blink = false
+
+	m.startTime = time.Now()
+	m.elapsed = 0
+	m.tokenCount = 0
+
+	return tea.Batch(
+		m.spinner.Tick,
+		func() tea.Msg {
+			resp, err := m.App.Prompt(m.App.Messages, m.chatReplyChunks, m.stopSignal)
+			if err != nil {
+				return msgChatResponseError{Err: err}
+			}
+			return msgChatResponse(*resp)
+		},
+	)
+}
--- a/pkg/tui/views/chat/input.go
+++ b/pkg/tui/views/chat/input.go
@ -0,0 +1,201 @@
+package chat
+
+import (
+	"fmt"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/model"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	tuiutil "git.mlow.ca/mlow/lmcli/pkg/tui/util"
+	tea "github.com/charmbracelet/bubbletea"
+)
+
+func (m *Model) handleInput(msg tea.KeyMsg) tea.Cmd {
+	switch m.focus {
+	case focusInput:
+		cmd := m.handleInputKey(msg)
+		if cmd != nil {
+			return cmd
+		}
+	case focusMessages:
+		cmd := m.handleMessagesKey(msg)
+		if cmd != nil {
+			return cmd
+		}
+	}
+
+	switch msg.String() {
+	case "esc":
+		if m.state == pendingResponse {
+			m.stopSignal <- struct{}{}
+			return shared.KeyHandled(msg)
+		}
+		return func() tea.Msg {
+			return shared.MsgViewChange(shared.ViewConversations)
+		}
+	case "ctrl+c":
+		if m.state == pendingResponse {
+			m.stopSignal <- struct{}{}
+			return shared.KeyHandled(msg)
+		}
+	case "ctrl+g":
+		if m.state == pendingResponse {
+			m.stopSignal <- struct{}{}
+			return shared.KeyHandled(msg)
+		}
+		return func() tea.Msg {
+			return shared.MsgViewChange(shared.ViewSettings)
+		}
+	case "ctrl+p":
+		m.persistence = !m.persistence
+		return shared.KeyHandled(msg)
+	case "ctrl+t":
+		m.showDetails = !m.showDetails
+		m.rebuildMessageCache()
+		m.updateContent()
+		return shared.KeyHandled(msg)
+	case "ctrl+w":
+		m.wrap = !m.wrap
+		m.rebuildMessageCache()
+		m.updateContent()
+		return shared.KeyHandled(msg)
+	case "ctrl+n":
+		m.App.NewConversation()
+		m.rebuildMessageCache()
+		m.updateContent()
+		return shared.KeyHandled(msg)
+	}
+	return nil
+}
+
+func (m *Model) scrollSelection(dir int) {
+	if m.selectedMessage+dir < 0 || m.selectedMessage+dir >= len(m.App.Messages) {
+		return
+	}
+
+	newIdx := m.selectedMessage
+	for i := newIdx + dir; i >= 0 && i < len(m.App.Messages); i += dir {
+		if !m.showDetails && m.App.Messages[i].Role.IsSystem() {
+			continue
+		}
+		newIdx = i
+		break
+	}
+	if newIdx != m.selectedMessage {
+		m.selectedMessage = newIdx
+		m.updateContent()
+	}
+	yOffset := m.messageOffsets[m.selectedMessage]
+	tuiutil.ScrollIntoView(&m.content, yOffset, m.content.Height/2)
+}
+
+// handleMessagesKey handles input when the messages pane is focused
+func (m *Model) handleMessagesKey(msg tea.KeyMsg) tea.Cmd {
+	switch msg.String() {
+	case "tab", "enter":
+		m.focus = focusInput
+		m.updateContent()
+		m.input.Focus()
+		return shared.KeyHandled(msg)
+	case "e":
+		if m.selectedMessage < len(m.App.Messages) {
+			m.editorTarget = selectedMessage
+			return tuiutil.OpenTempfileEditor(
+				"message.*.md",
+				m.App.Messages[m.selectedMessage].Content,
+				"# Edit the message below\n",
+			)
+		}
+		return nil
+	case "ctrl+k", "ctrl+up":
+		if m.selectedMessage > 0 {
+			m.scrollSelection(-1)
+		}
+		return shared.KeyHandled(msg)
+	case "ctrl+j", "ctrl+down":
+		if m.selectedMessage < len(m.App.Messages)-1 {
+			m.scrollSelection(1)
+		}
+		return shared.KeyHandled(msg)
+	case "ctrl+h", "ctrl+left", "ctrl+l", "ctrl+right":
+		dir := model.CyclePrev
+		if msg.String() == "ctrl+l" || msg.String() == "ctrl+right" {
+			dir = model.CycleNext
+		}
+
+		var cmd tea.Cmd
+		if m.selectedMessage == 0 {
+			cmd = m.cycleSelectedRoot(&m.App.Conversation, dir)
+		} else if m.selectedMessage > 0 {
+			cmd = m.cycleSelectedReply(&m.App.Messages[m.selectedMessage-1], dir)
+		}
+		return cmd
+	case "ctrl+r":
+		// prompt the model with all messages up to and including the selected message
+		if m.state == idle && m.selectedMessage < len(m.App.Messages) {
+			m.App.Messages = m.App.Messages[:m.selectedMessage+1]
+			m.messageCache = m.messageCache[:m.selectedMessage+1]
+			cmd := m.promptLLM()
+			m.updateContent()
+			m.content.GotoBottom()
+			return cmd
+		}
+	}
+	return nil
+}
+
+// handleInputKey handles input when the input textarea is focused
+func (m *Model) handleInputKey(msg tea.KeyMsg) tea.Cmd {
+	switch msg.String() {
+	case "esc":
+		m.focus = focusMessages
+		if len(m.App.Messages) > 0 {
+			if m.selectedMessage < 0 || m.selectedMessage >= len(m.App.Messages) {
+				m.selectedMessage = len(m.App.Messages) - 1
+			}
+			offset := m.messageOffsets[m.selectedMessage]
+			tuiutil.ScrollIntoView(&m.content, offset, m.content.Height/2)
+		}
+		m.updateContent()
+		m.input.Blur()
+		return shared.KeyHandled(msg)
+	case "ctrl+s":
+		if m.state != idle {
+			return nil
+		}
+
+		input := strings.TrimSpace(m.input.Value())
+		if input == "" {
+			return shared.KeyHandled(msg)
+		}
+
+		if len(m.App.Messages) > 0 && m.App.Messages[len(m.App.Messages)-1].Role.IsUser() {
+			return shared.WrapError(fmt.Errorf("Can't reply to a user message"))
+		}
+
+		m.addMessage(conversation.Message{
+			Role:    api.MessageRoleUser,
+			Content: input,
+		})
+
+		m.input.SetValue("")
+
+		var cmds []tea.Cmd
+		if m.persistence {
+			cmds = append(cmds, m.persistConversation())
+		}
+
+		cmds = append(cmds, m.promptLLM())
+
+		m.updateContent()
+		m.content.GotoBottom()
+		return tea.Batch(cmds...)
+	case "ctrl+e":
+		cmd := tuiutil.OpenTempfileEditor("message.*.md", m.input.Value(), "# Edit your input below\n")
+		m.editorTarget = input
+		return cmd
+	}
+	return nil
+}
--- a/pkg/tui/views/chat/update.go
+++ b/pkg/tui/views/chat/update.go
@ -0,0 +1,270 @@
+package chat
+
+import (
+	"strings"
+	"time"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	tuiutil "git.mlow.ca/mlow/lmcli/pkg/tui/util"
+	"github.com/charmbracelet/bubbles/cursor"
+	tea "github.com/charmbracelet/bubbletea"
+)
+
+func (m *Model) setMessage(i int, msg conversation.Message) {
+	if i >= len(m.App.Messages) {
+		panic("i out of range")
+	}
+	m.App.Messages[i] = msg
+	m.messageCache[i] = m.renderMessage(i)
+}
+
+func (m *Model) addMessage(msg conversation.Message) {
+	m.App.Messages = append(m.App.Messages, msg)
+	m.messageCache = append(m.messageCache, m.renderMessage(len(m.App.Messages)-1))
+}
+
+func (m *Model) setMessageContents(i int, content string) {
+	if i >= len(m.App.Messages) {
+		panic("i out of range")
+	}
+	m.App.Messages[i].Content = content
+	m.messageCache[i] = m.renderMessage(i)
+}
+
+func (m *Model) rebuildMessageCache() {
+	m.messageCache = make([]string, len(m.App.Messages))
+	for i := range m.App.Messages {
+		m.messageCache[i] = m.renderMessage(i)
+	}
+}
+
+func (m *Model) updateContent() {
+	atBottom := m.content.AtBottom()
+	m.content.SetContent(m.conversationMessagesView())
+	if atBottom {
+		m.content.GotoBottom()
+	}
+}
+
+func (m *Model) Update(msg tea.Msg) (shared.ViewModel, tea.Cmd) {
+	inputHandled := false
+
+	var cmds []tea.Cmd
+	switch msg := msg.(type) {
+	case tea.KeyMsg:
+		cmd := m.handleInput(msg)
+		if cmd != nil {
+			inputHandled = true
+			cmds = append(cmds, cmd)
+		}
+	case tea.WindowSizeMsg:
+		m.Width, m.Height = msg.Width, msg.Height
+		m.content.Width = msg.Width
+		m.input.SetWidth(msg.Width - m.input.FocusedStyle.Base.GetHorizontalFrameSize())
+		if len(m.App.Messages) > 0 {
+			m.rebuildMessageCache()
+			m.updateContent()
+		}
+	case shared.MsgViewEnter:
+		// wake up spinners and cursors
+		cmds = append(cmds, cursor.Blink, m.spinner.Tick)
+
+		// Refresh view
+		m.rebuildMessageCache()
+		m.updateContent()
+
+		if m.App.Conversation.ID > 0 {
+			// (re)load conversation contents
+			cmds = append(cmds, m.loadConversationMessages())
+		}
+	case tuiutil.MsgTempfileEditorClosed:
+		contents := string(msg)
+		switch m.editorTarget {
+		case input:
+			m.input.SetValue(contents)
+		case selectedMessage:
+			toEdit := m.App.Messages[m.selectedMessage]
+			if toEdit.Content != contents {
+				toEdit.Content = contents
+				m.setMessage(m.selectedMessage, toEdit)
+				if m.persistence && toEdit.ID > 0 {
+					// create clone of message with its new contents
+					cmds = append(cmds, m.cloneMessage(toEdit, true))
+				}
+			}
+		}
+	case msgConversationMessagesLoaded:
+		m.App.Messages = msg.messages
+		if m.selectedMessage == -1 {
+			m.selectedMessage = len(msg.messages) - 1
+		} else {
+			m.selectedMessage = min(m.selectedMessage, len(m.App.Messages))
+		}
+		m.rebuildMessageCache()
+		m.updateContent()
+	case msgChatResponseChunk:
+		cmds = append(cmds, m.waitForResponseChunk()) // wait for the next chunk
+
+		if msg.Content == "" {
+			break
+		}
+
+		last := len(m.App.Messages) - 1
+		if last >= 0 && m.App.Messages[last].Role.IsAssistant() {
+			// append chunk to existing message
+			m.setMessageContents(last, m.App.Messages[last].Content+msg.Content)
+		} else {
+			// use chunk in a new message
+			m.addMessage(conversation.Message{
+				Role:    api.MessageRoleAssistant,
+				Content: msg.Content,
+			})
+		}
+		m.updateContent()
+
+		// show cursor and reset blink interval (simulate typing)
+		m.replyCursor.Blink = false
+		cmds = append(cmds, m.replyCursor.BlinkCmd())
+
+		m.tokenCount += msg.TokenCount
+		m.elapsed = time.Now().Sub(m.startTime)
+	case msgChatResponse:
+		m.state = idle
+
+		reply := conversation.Message(msg)
+		reply.Content = strings.TrimSpace(reply.Content)
+
+		last := len(m.App.Messages) - 1
+		if last < 0 {
+			panic("Unexpected empty messages handling msgAssistantReply")
+		}
+
+		if m.App.Messages[last].Role.IsAssistant() {
+			// TODO: handle continuations gracefully - only some models support them
+			m.setMessage(last, reply)
+		} else {
+			m.addMessage(reply)
+		}
+
+		if reply.Role == api.MessageRoleToolCall {
+			// TODO: user confirmation before execution
+			// m.state = confirmToolUse
+			cmds = append(cmds, m.executeToolCalls(reply.ToolCalls))
+		}
+
+		if m.persistence {
+			cmds = append(cmds, m.persistConversation())
+		}
+
+		if m.App.Conversation.Title == "" && len(m.App.Messages) > 0 {
+			cmds = append(cmds, m.generateConversationTitle())
+		}
+	case msgChatResponseCanceled:
+		m.state = idle
+		m.updateContent()
+	case msgChatResponseError:
+		m.state = idle
+		m.updateContent()
+		return m, shared.WrapError(msg.Err)
+	case msgToolResults:
+		last := len(m.App.Messages) - 1
+		if last < 0 {
+			panic("Unexpected empty messages handling msgAssistantReply")
+		}
+
+		if m.App.Messages[last].Role != api.MessageRoleToolCall {
+			panic("Previous message not a tool call, unexpected")
+		}
+
+		m.addMessage(conversation.Message{
+			Role:        api.MessageRoleToolResult,
+			ToolResults: conversation.ToolResults(msg),
+		})
+
+		if m.persistence {
+			cmds = append(cmds, m.persistConversation())
+		}
+
+		m.updateContent()
+	case msgConversationTitleGenerated:
+		title := string(msg)
+		m.App.Conversation.Title = title
+		if m.persistence && m.App.Conversation.ID > 0 {
+			cmds = append(cmds, m.persistConversation())
+		}
+	case cursor.BlinkMsg:
+		if m.state == pendingResponse {
+			// ensure we show the updated "wait for response" cursor blink state
+			last := len(m.App.Messages) - 1
+			m.messageCache[last] = m.renderMessage(last)
+			m.updateContent()
+		}
+	case msgConversationPersisted:
+		m.App.Conversation = conversation.Conversation(msg)
+		cmds = append(cmds, m.persistMessages())
+	case msgMessagesPersisted:
+		m.App.Messages = msg
+		m.rebuildMessageCache()
+		m.updateContent()
+	case msgMessageCloned:
+		cmds = append(cmds, m.loadConversationMessages())
+	case msgSelectedRootCycled, msgSelectedReplyCycled, msgMessageUpdated:
+		cmds = append(cmds, m.loadConversationMessages())
+	}
+
+	var cmd tea.Cmd
+	m.spinner, cmd = m.spinner.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	m.replyCursor, cmd = m.replyCursor.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+
+	prevInputLineCnt := m.input.LineCount()
+
+	if !inputHandled {
+		m.input, cmd = m.input.Update(msg)
+		if cmd != nil {
+			inputHandled = true
+			cmds = append(cmds, cmd)
+		}
+	}
+
+	if !inputHandled {
+		m.content, cmd = m.content.Update(msg)
+		if cmd != nil {
+			cmds = append(cmds, cmd)
+		}
+	}
+
+	// this is a pretty nasty hack to ensure the input area viewport doesn't
+	// scroll below its content, which can happen when the input viewport
+	// height has grown, or previously entered lines have been deleted
+	if prevInputLineCnt != m.input.LineCount() {
+		// dist is the distance we'd need to scroll up from the current cursor
+		// position to position the last input line at the bottom of the
+		// viewport. if negative, we're already scrolled above the bottom
+		dist := m.input.Line() - (m.input.LineCount() - m.input.Height())
+		if dist > 0 {
+			for i := 0; i < dist; i++ {
+				// move cursor up until content reaches the bottom of the viewport
+				m.input.CursorUp()
+			}
+			m.input, _ = m.input.Update(nil)
+			for i := 0; i < dist; i++ {
+				// move cursor back down to its previous position
+				m.input.CursorDown()
+			}
+			m.input, _ = m.input.Update(nil)
+		}
+	}
+
+	if len(cmds) > 0 {
+		return m, tea.Batch(cmds...)
+	}
+	return m, nil
+}
--- a/pkg/tui/views/chat/view.go
+++ b/pkg/tui/views/chat/view.go
@ -0,0 +1,365 @@
+package chat
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/api"
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/styles"
+	tuiutil "git.mlow.ca/mlow/lmcli/pkg/tui/util"
+	"github.com/charmbracelet/lipgloss"
+	"github.com/muesli/reflow/wordwrap"
+	"github.com/muesli/reflow/wrap"
+	"gopkg.in/yaml.v3"
+)
+
+// styles
+var (
+	boldStyle      = lipgloss.NewStyle().Bold(true)
+	faintStyle     = lipgloss.NewStyle().Faint(true)
+	boldFaintStyle = lipgloss.NewStyle().Faint(true).Bold(true)
+
+	messageHeadingStyle = lipgloss.NewStyle().
+				MarginTop(1).
+				MarginBottom(1)
+
+	userStyle = boldFaintStyle.Foreground(lipgloss.Color("10"))
+
+	assistantStyle = boldFaintStyle.Foreground(lipgloss.Color("12"))
+
+	systemStyle = boldStyle.Foreground(lipgloss.Color("8"))
+
+	messageStyle = lipgloss.NewStyle().
+			PaddingLeft(2).
+			PaddingRight(2)
+
+	inputFocusedStyle = lipgloss.NewStyle().
+				Border(lipgloss.RoundedBorder(), true, true, true, false)
+
+	inputBlurredStyle = lipgloss.NewStyle().
+				Faint(true).
+				Border(lipgloss.RoundedBorder(), true, true, true, false)
+
+	footerStyle = lipgloss.NewStyle().Padding(0, 1)
+)
+
+func (m *Model) renderMessageHeading(i int, message *conversation.Message) string {
+	friendly := message.Role.FriendlyRole()
+	style := systemStyle
+
+	switch message.Role {
+	case api.MessageRoleUser:
+		style = userStyle
+	case api.MessageRoleAssistant:
+		style = assistantStyle
+	case api.MessageRoleToolCall:
+		style = assistantStyle
+		friendly = api.MessageRoleAssistant.FriendlyRole()
+	case api.MessageRoleSystem:
+	case api.MessageRoleToolResult:
+	}
+
+	user := style.Render(friendly)
+
+	var prefix, suffix string
+
+	if i == m.selectedMessage && m.focus == focusMessages {
+		prefix = "> "
+	} else {
+		prefix = " "
+	}
+
+	if i == 0 && m.App.Conversation.SelectedRootID != nil && len(m.App.Conversation.RootMessages) > 1 {
+		selectedRootIndex := 0
+		for j, reply := range m.App.Conversation.RootMessages {
+			if reply.ID == *m.App.Conversation.SelectedRootID {
+				selectedRootIndex = j
+				break
+			}
+		}
+		suffix += faintStyle.Render(fmt.Sprintf(" <%d/%d>", selectedRootIndex+1, len(m.App.Conversation.RootMessages)))
+	}
+	if i > 0 && len(m.App.Messages[i-1].Replies) > 1 {
+		// Find the selected reply index
+		selectedReplyIndex := 0
+		for j, reply := range m.App.Messages[i-1].Replies {
+			if reply.ID == *m.App.Messages[i-1].SelectedReplyID {
+				selectedReplyIndex = j
+				break
+			}
+		}
+		suffix += faintStyle.Render(fmt.Sprintf(" <%d/%d>", selectedReplyIndex+1, len(m.App.Messages[i-1].Replies)))
+	}
+
+	if message.ID == 0 {
+		suffix += faintStyle.Render(" (not saved)")
+	}
+
+	heading := prefix + user + suffix
+
+	if message.Metadata.GenerationModel != nil && m.showDetails {
+		heading += faintStyle.Render(
+			fmt.Sprintf(" | %s", *message.Metadata.GenerationModel),
+		)
+	}
+
+	return messageHeadingStyle.Render(heading)
+}
+
+// renderMessages renders the message at the given index as it should be shown
+// *at this moment* - we render differently depending on the current application
+// state (window size, etc, etc).
+func (m *Model) renderMessage(i int) string {
+	msg := &m.App.Messages[i]
+
+	// Write message contents
+	sb := &strings.Builder{}
+	sb.Grow(len(msg.Content) * 2)
+	if msg.Content != "" {
+		err := m.App.Ctx.Chroma.Highlight(sb, msg.Content)
+		if err != nil {
+			sb.Reset()
+			sb.WriteString(msg.Content)
+		}
+	}
+
+	isLast := i == len(m.App.Messages)-1
+	isAssistant := msg.Role == api.MessageRoleAssistant
+
+	if m.state == pendingResponse && isLast && isAssistant {
+		// Show the assistant's cursor
+		sb.WriteString(m.replyCursor.View())
+	}
+
+	// Write tool call info
+	var toolString string
+	switch msg.Role {
+	case api.MessageRoleToolCall:
+		bytes, err := yaml.Marshal(msg.ToolCalls)
+		if err != nil {
+			toolString = "Could not serialize ToolCalls"
+		} else {
+			toolString = "tool_calls:\n" + string(bytes)
+		}
+	case api.MessageRoleToolResult:
+		type renderedResult struct {
+			ToolName string `yaml:"tool"`
+			Result   any    `yaml:"result,omitempty"`
+		}
+
+		var toolResults []renderedResult
+		for _, result := range msg.ToolResults {
+			if m.showDetails {
+				var jsonResult interface{}
+				err := json.Unmarshal([]byte(result.Result), &jsonResult)
+				if err != nil {
+					// If parsing as JSON fails, treat Result as a plain string
+					toolResults = append(toolResults, renderedResult{
+						ToolName: result.ToolName,
+						Result:   result.Result,
+					})
+				} else {
+					// If parsing as JSON succeeds, marshal the parsed JSON into YAML
+					toolResults = append(toolResults, renderedResult{
+						ToolName: result.ToolName,
+						Result:   &jsonResult,
+					})
+				}
+			} else {
+				// Only show the tool name when results are hidden
+				toolResults = append(toolResults, renderedResult{
+					ToolName: result.ToolName,
+					Result:   "(hidden, press ctrl+t to view)",
+				})
+			}
+		}
+
+		bytes, err := yaml.Marshal(toolResults)
+		if err != nil {
+			toolString = "Could not serialize ToolResults"
+		} else {
+			toolString = "tool_results:\n" + string(bytes)
+		}
+	}
+
+	if toolString != "" {
+		toolString = strings.TrimRight(toolString, "\n")
+		if msg.Content != "" {
+			sb.WriteString("\n\n")
+		}
+		_ = m.App.Ctx.Chroma.HighlightLang(sb, toolString, "yaml")
+	}
+
+	content := strings.TrimRight(sb.String(), "\n")
+
+	if m.wrap {
+		wrapWidth := m.content.Width - messageStyle.GetHorizontalPadding()
+		// first we word-wrap text to slightly less than desired width (since
+		// wordwrap seems to have an off-by-1 issue), then hard wrap at
+		// desired with
+		content = wrap.String(wordwrap.String(content, wrapWidth-2), wrapWidth)
+	}
+
+	return messageStyle.Width(0).Render(content)
+}
+
+// render the conversation into a string
+func (m *Model) conversationMessagesView() string {
+	m.messageOffsets = make([]int, len(m.App.Messages))
+	lineCnt := 1
+
+	sb := strings.Builder{}
+	for i, message := range m.App.Messages {
+		m.messageOffsets[i] = lineCnt
+
+		if !m.showDetails && message.Role.IsSystem() {
+			continue
+		}
+
+		heading := m.renderMessageHeading(i, &message)
+		sb.WriteString(heading)
+		sb.WriteString("\n")
+		lineCnt += lipgloss.Height(heading)
+
+		rendered := m.messageCache[i]
+		sb.WriteString(rendered)
+		sb.WriteString("\n")
+		lineCnt += lipgloss.Height(rendered)
+	}
+
+	// Render a placeholder for the incoming assistant reply
+	if m.state == pendingResponse && m.App.Messages[len(m.App.Messages)-1].Role != api.MessageRoleAssistant {
+		heading := m.renderMessageHeading(-1, &conversation.Message{
+			Role: api.MessageRoleAssistant,
+			Metadata: conversation.MessageMeta{
+				GenerationModel: &m.App.Model,
+			},
+		})
+		sb.WriteString(heading)
+		sb.WriteString("\n")
+		sb.WriteString(messageStyle.Width(0).Render(m.replyCursor.View()))
+		sb.WriteString("\n")
+	}
+	return sb.String()
+}
+
+func (m *Model) Content(width, height int) string {
+	// calculate clamped input height to accomodate input text
+	// minimum 4 lines, maximum half of content area
+	inputHeight := max(4, min(height/2, m.input.LineCount()))
+	m.input.SetHeight(inputHeight)
+	input := m.input.View()
+
+	// remaining height towards content
+	m.content.Width, m.content.Height = width, height-tuiutil.Height(input)
+	content := m.content.View()
+	return lipgloss.JoinVertical(lipgloss.Left, content, input)
+}
+
+func (m *Model) Header(width int) string {
+	titleStyle := lipgloss.NewStyle().Bold(true)
+	var title string
+	if m.App.Conversation.Title != "" {
+		title = m.App.Conversation.Title
+	} else {
+		title = "Untitled"
+	}
+	title = tuiutil.TruncateRightToCellWidth(title, width-styles.Header.GetHorizontalPadding(), "...")
+	header := titleStyle.Render(title)
+	return styles.Header.Width(width).Render(header)
+}
+
+func (m *Model) Footer(width int) string {
+	segmentStyle := lipgloss.NewStyle().Faint(true)
+	segmentSeparator := segmentStyle.Render(" | ")
+
+	// Left segments
+	leftSegments := make([]string, 0, 4)
+
+	if m.state == pendingResponse {
+		leftSegments = append(leftSegments, segmentStyle.Render(m.spinner.View()))
+	} else {
+		leftSegments = append(leftSegments, segmentStyle.Render("∙∙∙"))
+	}
+
+	if m.elapsed > 0 && m.tokenCount > 0 {
+		throughput := fmt.Sprintf("%.0f t/sec", float64(m.tokenCount)/m.elapsed.Seconds())
+		leftSegments = append(leftSegments, segmentStyle.Render(throughput))
+	}
+
+	// var status string
+	// switch m.state {
+	// case pendingResponse:
+	// 	status = "Press ctrl+c to cancel"
+	// default:
+	// 	status = "Press ctrl+s to send"
+	// }
+	// leftSegments = append(leftSegments, segmentStyle.Render(status))
+
+	// Right segments
+	rightSegments := make([]string, 0, 8)
+
+	if m.App.Agent != nil {
+		rightSegments = append(rightSegments, segmentStyle.Render(m.App.Agent.Name))
+	}
+
+	model := segmentStyle.Render(m.App.ActiveModel(lipgloss.NewStyle()))
+	rightSegments = append(rightSegments, model)
+
+	savingStyle := segmentStyle.Bold(true)
+	saving := ""
+	if m.persistence {
+		saving = savingStyle.Foreground(lipgloss.Color("2")).Render("💾✅")
+	} else {
+		saving = savingStyle.Foreground(lipgloss.Color("1")).Render("💾❌")
+	}
+	rightSegments = append(rightSegments, saving)
+
+	return m.layoutFooter(width, leftSegments, rightSegments, segmentSeparator)
+}
+
+func (m *Model) layoutFooter(
+	width int,
+	leftSegments []string,
+	rightSegments []string,
+	segmentSeparator string,
+) string {
+	left := strings.Join(leftSegments, segmentSeparator)
+	right := strings.Join(rightSegments, segmentSeparator)
+
+	leftWidth := tuiutil.Width(left)
+	rightWidth := tuiutil.Width(right)
+	sepWidth := tuiutil.Width(segmentSeparator)
+	frameWidth := footerStyle.GetHorizontalFrameSize()
+
+	availableWidth := width - frameWidth - leftWidth - rightWidth
+
+	if availableWidth >= sepWidth {
+		// Everything fits
+		padding := strings.Repeat(" ", availableWidth)
+		return footerStyle.Render(left + padding + right)
+	}
+
+	// Inserted between left and right segments when they're being truncated
+	div := "..."
+
+	totalAvailableWidth := width - frameWidth
+	availableTruncWidth := totalAvailableWidth - len(div)
+
+	minVisibleLength := 3
+	if availableTruncWidth < 2*minVisibleLength {
+		minVisibleLength = availableTruncWidth / 2
+	}
+
+	leftProportion := float64(leftWidth) / float64(leftWidth+rightWidth)
+
+	newLeftWidth := int(max(float64(minVisibleLength), leftProportion*float64(availableTruncWidth)))
+	newRightWidth := totalAvailableWidth - newLeftWidth
+
+	truncatedLeft := faintStyle.Render(tuiutil.TruncateRightToCellWidth(left, newLeftWidth, ""))
+	truncatedRight := faintStyle.Render(tuiutil.TruncateLeftToCellWidth(right, newRightWidth, "..."))
+
+	return footerStyle.Width(width).Render(truncatedLeft + truncatedRight)
+}
--- a/pkg/tui/views/conversations/conversations.go
+++ b/pkg/tui/views/conversations/conversations.go
@ -0,0 +1,326 @@
+package conversations
+
+import (
+	"fmt"
+	"strings"
+	"time"
+
+	"git.mlow.ca/mlow/lmcli/pkg/conversation"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/bubbles"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/model"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/styles"
+	tuiutil "git.mlow.ca/mlow/lmcli/pkg/tui/util"
+	"git.mlow.ca/mlow/lmcli/pkg/util"
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+type (
+	// sent when conversation list is loaded
+	msgConversationsLoaded conversation.ConversationList
+	// sent when a single conversation is loaded
+	msgConversationLoaded conversation.Conversation
+	// sent when a conversation is deleted
+	msgConversationDeleted struct{}
+)
+
+type Model struct {
+	App    *model.AppModel
+	width  int
+	height int
+
+	cursor      int
+	itemOffsets []int // conversation y offsets
+
+	content       viewport.Model
+	confirmPrompt bubbles.ConfirmPrompt
+}
+
+func Conversations(app *model.AppModel) *Model {
+	viewport.New(0, 0)
+	m := Model{
+		App:     app,
+		content: viewport.New(0, 0),
+	}
+	return &m
+}
+
+func (m *Model) handleInput(msg tea.KeyMsg) tea.Cmd {
+	if m.confirmPrompt.Focused() {
+		var cmd tea.Cmd
+		m.confirmPrompt, cmd = m.confirmPrompt.Update(msg)
+		if cmd != nil {
+			return cmd
+		}
+	}
+
+	conversations := m.App.Conversations.Items
+
+	switch msg.String() {
+	case "enter":
+		if len(conversations) > 0 && m.cursor < len(conversations) {
+			return m.loadConversation(conversations[m.cursor].ID)
+		}
+	case "j", "down":
+		if m.cursor < len(conversations)-1 {
+			m.cursor++
+			if m.cursor == len(conversations)-1 {
+				m.content.GotoBottom()
+			} else {
+				// this hack positions the *next* conversatoin slightly
+				// *off* the screen, ensuring the entire m.cursor is shown,
+				// even if its height may not be constant due to wrapping.
+				tuiutil.ScrollIntoView(&m.content, m.itemOffsets[m.cursor+1], -1)
+			}
+			m.content.SetContent(m.renderConversationList())
+		} else {
+			m.cursor = len(conversations) - 1
+			m.content.GotoBottom()
+		}
+		return shared.KeyHandled(msg)
+	case "k", "up":
+		if m.cursor > 0 {
+			m.cursor--
+			if m.cursor == 0 {
+				m.content.GotoTop()
+			} else {
+				tuiutil.ScrollIntoView(&m.content, m.itemOffsets[m.cursor], 1)
+			}
+			m.content.SetContent(m.renderConversationList())
+		} else {
+			m.cursor = 0
+			m.content.GotoTop()
+		}
+		return shared.KeyHandled(msg)
+	case "n":
+		m.App.NewConversation()
+		return shared.ChangeView(shared.ViewChat)
+	case "d":
+		if !m.confirmPrompt.Focused() && len(conversations) > 0 && m.cursor < len(conversations) {
+			title := conversations[m.cursor].Title
+			if title == "" {
+				title = "(untitled)"
+			}
+			m.confirmPrompt = bubbles.NewConfirmPrompt(
+				fmt.Sprintf("Delete '%s'?", title),
+				conversations[m.cursor],
+			)
+			m.confirmPrompt.Style = lipgloss.NewStyle().
+				Bold(true).
+				Foreground(lipgloss.Color("3"))
+			return shared.KeyHandled(msg)
+		}
+	case "c":
+		// copy/clone conversation
+	case "r":
+		// show prompt to rename conversation
+	case "shift+r":
+		// show prompt to generate name for conversation
+	}
+	return nil
+}
+
+func (m *Model) Init() tea.Cmd {
+	return nil
+}
+
+func (m *Model) Update(msg tea.Msg) (shared.ViewModel, tea.Cmd) {
+	isInput := false
+	inputHandled := false
+
+	var cmds []tea.Cmd
+	switch msg := msg.(type) {
+	case tea.KeyMsg:
+		isInput = true
+		cmd := m.handleInput(msg)
+		if cmd != nil {
+			cmds = append(cmds, cmd)
+			inputHandled = true
+		}
+	case shared.MsgViewEnter:
+		cmds = append(cmds, m.loadConversations())
+		m.content.SetContent(m.renderConversationList())
+	case tea.WindowSizeMsg:
+		m.width, m.height = msg.Width, msg.Height
+		m.content.SetContent(m.renderConversationList())
+	case msgConversationsLoaded:
+		m.App.Conversations = conversation.ConversationList(msg)
+		m.cursor = max(0, min(len(m.App.Conversations.Items), m.cursor))
+		m.content.SetContent(m.renderConversationList())
+	case msgConversationLoaded:
+		m.App.ClearConversation()
+		m.App.Conversation = conversation.Conversation(msg)
+		cmds = append(cmds, func() tea.Msg {
+			return shared.MsgViewChange(shared.ViewChat)
+		})
+	case bubbles.MsgConfirmPromptAnswered:
+		m.confirmPrompt.Blur()
+		if msg.Value {
+			conv, ok := msg.Payload.(conversation.ConversationListItem)
+			if ok {
+				cmds = append(cmds, m.deleteConversation(conv))
+			}
+		}
+	case msgConversationDeleted:
+		cmds = append(cmds, m.loadConversations())
+	}
+
+	if !isInput || !inputHandled {
+		content, cmd := m.content.Update(msg)
+		m.content = content
+		if cmd != nil {
+			cmds = append(cmds, cmd)
+		}
+	}
+
+	if len(cmds) > 0 {
+		return m, tea.Batch(cmds...)
+	}
+
+	return m, nil
+}
+
+func (m *Model) loadConversations() tea.Cmd {
+	return func() tea.Msg {
+		list, err := m.App.Ctx.Conversations.LoadConversationList()
+		if err != nil {
+			return shared.AsMsgError(fmt.Errorf("Could not load conversations: %v", err))
+		}
+		return msgConversationsLoaded(list)
+	}
+}
+
+func (m *Model) loadConversation(conversationID uint) tea.Cmd {
+	return func() tea.Msg {
+		conversation, err := m.App.Ctx.Conversations.GetConversationByID(conversationID)
+		if err != nil {
+			return shared.AsMsgError(fmt.Errorf("Could not load conversation %d: %v", conversationID, err))
+		}
+		return msgConversationLoaded(*conversation)
+	}
+}
+
+func (m *Model) deleteConversation(conv conversation.ConversationListItem) tea.Cmd {
+	return func() tea.Msg {
+		err := m.App.Ctx.Conversations.DeleteConversationById(conv.ID)
+		if err != nil {
+			return shared.AsMsgError(fmt.Errorf("Could not delete conversation: %v", err))
+		}
+		return msgConversationDeleted{}
+	}
+}
+
+func (m *Model) Header(width int) string {
+	titleStyle := lipgloss.NewStyle().Bold(true)
+	header := titleStyle.Render("Conversations")
+	return styles.Header.Width(width).Render(header)
+}
+
+func (m *Model) Content(width int, height int) string {
+	m.content.Width, m.content.Height = width, height
+	return m.content.View()
+}
+
+func (m *Model) Footer(width int) string {
+	if m.confirmPrompt.Focused() {
+		return lipgloss.NewStyle().Width(width).Render(m.confirmPrompt.View())
+	}
+	return ""
+}
+
+func (m *Model) renderConversationList() string {
+	now := time.Now()
+	midnight := time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, now.Location())
+	monthStart := time.Date(now.Year(), now.Month(), 1, 0, 0, 0, 0, now.Location())
+	dayOfWeek := int(now.Weekday())
+	categories := []struct {
+		name   string
+		cutoff time.Duration
+	}{
+		{"Today", now.Sub(midnight)},
+		{"Yesterday", now.Sub(midnight.AddDate(0, 0, -1))},
+		{"This week", now.Sub(midnight.AddDate(0, 0, -dayOfWeek))},
+		{"Last week", now.Sub(midnight.AddDate(0, 0, -(dayOfWeek + 7)))},
+		{"This month", now.Sub(monthStart)},
+		{"Last month", now.Sub(monthStart.AddDate(0, -1, 0))},
+		{"2 Months ago", now.Sub(monthStart.AddDate(0, -2, 0))},
+		{"3 Months ago", now.Sub(monthStart.AddDate(0, -3, 0))},
+		{"4 Months ago", now.Sub(monthStart.AddDate(0, -4, 0))},
+		{"5 Months ago", now.Sub(monthStart.AddDate(0, -5, 0))},
+		{"6 Months ago", now.Sub(monthStart.AddDate(0, -6, 0))},
+		{"Older", now.Sub(time.Time{})},
+	}
+
+	categoryStyle := lipgloss.NewStyle().
+		MarginBottom(1).
+		Foreground(lipgloss.Color("12")).
+		PaddingLeft(1).
+		Bold(true)
+
+	itemStyle := lipgloss.NewStyle().
+		MarginBottom(1)
+
+	ageStyle := lipgloss.NewStyle().Faint(true).SetString()
+	titleStyle := lipgloss.NewStyle().Bold(true)
+	untitledStyle := lipgloss.NewStyle().Faint(true).Italic(true)
+	selectedStyle := lipgloss.NewStyle().Faint(true).Foreground(lipgloss.Color("6"))
+
+	var (
+		currentOffset   int
+		currentCategory string
+		sb              strings.Builder
+	)
+
+	m.itemOffsets = make([]int, len(m.App.Conversations.Items))
+	sb.WriteRune('\n')
+	currentOffset += 1
+
+	for i, c := range m.App.Conversations.Items {
+		lastReplyAge := now.Sub(c.LastMessageAt)
+
+		var category string
+		for _, g := range categories {
+			if lastReplyAge < g.cutoff {
+				category = g.name
+				break
+			}
+		}
+
+		// print the category
+		if category != currentCategory {
+			currentCategory = category
+			heading := categoryStyle.Render(currentCategory)
+			sb.WriteString(heading)
+			currentOffset += tuiutil.Height(heading)
+			sb.WriteRune('\n')
+		}
+
+		tStyle := titleStyle
+		if c.Title == "" {
+			tStyle = tStyle.Inherit(untitledStyle).SetString("(untitled)")
+		}
+		if i == m.cursor {
+			tStyle = tStyle.Inherit(selectedStyle)
+		}
+
+		title := tStyle.Width(m.width - 3).PaddingLeft(2).Render(c.Title)
+		if i == m.cursor {
+			title = ">" + title[1:]
+		}
+
+		m.itemOffsets[i] = currentOffset
+		item := itemStyle.Render(fmt.Sprintf(
+			"%s\n  %s",
+			title,
+			ageStyle.Render(util.HumanTimeElapsedSince(lastReplyAge)),
+		))
+		sb.WriteString(item)
+		currentOffset += tuiutil.Height(item)
+		if i < len(m.App.Conversations.Items)-1 {
+			sb.WriteRune('\n')
+		}
+	}
+	return sb.String()
+}
--- a/pkg/tui/views/settings/settings.go
+++ b/pkg/tui/views/settings/settings.go
@ -0,0 +1,137 @@
+package settings
+
+import (
+	"strings"
+
+	"git.mlow.ca/mlow/lmcli/pkg/tui/bubbles/list"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/model"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/shared"
+	"git.mlow.ca/mlow/lmcli/pkg/tui/styles"
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+type Model struct {
+	App       *model.AppModel
+	prevView  shared.View
+	content   viewport.Model
+	modelList list.Model
+	width     int
+	height    int
+}
+
+type modelOpt struct {
+	provider string
+	model    string
+}
+
+const (
+	modelListId int = iota + 1
+)
+
+func Settings(app *model.AppModel) *Model {
+	m := &Model{
+		App:     app,
+		content: viewport.New(0, 0),
+	}
+	return m
+}
+
+func (m *Model) Init() tea.Cmd {
+	m.modelList = list.NewWithGroups(m.getModelOptions())
+	m.modelList.ID = modelListId
+	return nil
+}
+
+func (m *Model) Update(msg tea.Msg) (shared.ViewModel, tea.Cmd) {
+	var cmd tea.Cmd
+
+	switch msg := msg.(type) {
+	case tea.KeyMsg:
+		m.modelList, cmd = m.modelList.Update(msg)
+		if cmd != nil {
+			return m, cmd
+		}
+
+		switch msg.String() {
+		case "esc":
+			return m, func() tea.Msg {
+				return shared.MsgViewChange(m.prevView)
+			}
+		}
+	case shared.MsgViewEnter:
+		m.prevView = shared.View(msg)
+		m.modelList.Focus()
+		m.content.SetContent(m.renderContent())
+	case tea.WindowSizeMsg:
+		m.width, m.height = msg.Width, msg.Height
+		m.content.Width = msg.Width
+		m.content.Height = msg.Height
+		m.content.SetContent(m.renderContent())
+	case list.MsgOptionSelected:
+		switch msg.ID {
+		case modelListId:
+			if modelOpt, ok := msg.Option.Value.(modelOpt); ok {
+				m.App.Model = modelOpt.model
+				m.App.ProviderName = modelOpt.provider
+			}
+			return m, shared.ChangeView(m.prevView)
+		}
+	}
+
+	m.modelList, cmd = m.modelList.Update(msg)
+	if cmd != nil {
+		return m, cmd
+	}
+
+	m.content.SetContent(m.renderContent())
+	return m, nil
+}
+
+func (m *Model) getModelOptions() []list.OptionGroup {
+	modelOpts := []list.OptionGroup{}
+	for _, p := range m.App.Ctx.Config.Providers {
+		provider := p.Name
+		if provider == "" {
+			provider = p.Kind
+		}
+		providerLabel := p.Display
+		if providerLabel == "" {
+			providerLabel = strings.ToUpper(provider[:1]) + provider[1:]
+		}
+		group := list.OptionGroup{
+			Name: providerLabel,
+		}
+		for _, model := range p.Models {
+			group.Options = append(group.Options, list.Option{
+				Label: model,
+				Value: modelOpt{provider, model},
+			})
+		}
+		modelOpts = append(modelOpts, group)
+	}
+	return modelOpts
+}
+
+func (m *Model) Header(width int) string {
+	boldStyle := lipgloss.NewStyle().Bold(true)
+	// TODO: update header depending on active settings mode (model, agent, etc)
+	header := boldStyle.Render("Model selection")
+	return styles.Header.Width(width).Render(header)
+}
+
+func (m *Model) Content(width, height int) string {
+	// TODO: see Header()
+	currentModel := " Active model: " + m.App.ActiveModel(lipgloss.NewStyle())
+	m.modelList.Width, m.modelList.Height = width, height - 2
+	return "\n" + currentModel + "\n" + m.modelList.View()
+}
+
+func (m *Model) Footer(width int) string {
+	return ""
+}
+
+func (m *Model) renderContent() string {
+	return m.modelList.View()
+}
--- a/pkg/util/tty/highlight.go
+++ b/pkg/util/tty/highlight.go
@ -58,3 +58,29 @@ func (s *ChromaHighlighter) HighlightS(text string) (string, error) {
 	s.formatter.Format(&sb, s.style, it)
 	return sb.String(), nil
 }
+
+func (s *ChromaHighlighter) HighlightLang(w io.Writer, text string, lang string) (error) {
+	l := lexers.Get(lang)
+	if l == nil {
+		l = lexers.Fallback
+	}
+	l = chroma.Coalesce(l)
+	old := s.lexer
+	s.lexer = l
+	err := s.Highlight(w, text)
+	s.lexer = old
+	return err
+}
+
+func (s *ChromaHighlighter) HighlightLangS(text string, lang string) (string, error) {
+	l := lexers.Get(lang)
+	if l == nil {
+		l = lexers.Fallback
+	}
+	l = chroma.Coalesce(l)
+	old := s.lexer
+	s.lexer = l
+	highlighted, err := s.HighlightS(text)
+	s.lexer = old
+	return highlighted, err
+}
--- a/pkg/util/util.go
+++ b/pkg/util/util.go
@ -21,7 +21,7 @@ func InputFromEditor(placeholder string, pattern string, content string) (string
 	msgFile, _ := os.CreateTemp("/tmp", pattern)
 	defer os.Remove(msgFile.Name())

-	os.WriteFile(msgFile.Name(), []byte(placeholder + content), os.ModeAppend)
+	os.WriteFile(msgFile.Name(), []byte(placeholder+content), os.ModeAppend)

 	editor := os.Getenv("EDITOR")
 	if editor == "" {
@ -137,8 +137,8 @@ func SetStructDefaults(data interface{}) bool {
 		}

 		// Get the "default" struct tag
-		defaultTag := v.Type().Field(i).Tag.Get("default")
-		if defaultTag == "" {
+		defaultTag, ok := v.Type().Field(i).Tag.Lookup("default")
+		if !ok {
 			continue
 		}

@ -147,16 +147,16 @@ func SetStructDefaults(data interface{}) bool {
 		case reflect.String:
 			defaultValue := defaultTag
 			field.Set(reflect.ValueOf(&defaultValue))
+		case reflect.Uint, reflect.Uint32, reflect.Uint64:
+			intValue, _ := strconv.ParseUint(defaultTag, 10, e.Bits())
+			field.Set(reflect.New(e))
+			field.Elem().SetUint(intValue)
 		case reflect.Int, reflect.Int32, reflect.Int64:
-			intValue, _ := strconv.ParseInt(defaultTag, 10, 64)
+			intValue, _ := strconv.ParseInt(defaultTag, 10, e.Bits())
 			field.Set(reflect.New(e))
 			field.Elem().SetInt(intValue)
-		case reflect.Float32:
-			floatValue, _ := strconv.ParseFloat(defaultTag, 32)
-			field.Set(reflect.New(e))
-			field.Elem().SetFloat(floatValue)
-		case reflect.Float64:
-			floatValue, _ := strconv.ParseFloat(defaultTag, 64)
+		case reflect.Float32, reflect.Float64:
+			floatValue, _ := strconv.ParseFloat(defaultTag, e.Bits())
 			field.Set(reflect.New(e))
 			field.Elem().SetFloat(floatValue)
 		case reflect.Bool:
Author	SHA1	Message	Date
Matt Low	f05e2e30f7	Update TODO.md	2024-10-27 18:03:15 +00:00
Matt Low	ec21a02ec0	Tweaks/cleanups to conversation management in tui - Pass around message/conversation values instead of pointers where it makes more sense, and store values instead of pointers in the globally (within the TUI) shared `App` (pointers provide no utility here). - Split conversation persistence into separate conversation/message saving stages	2024-10-25 16:57:15 +00:00
Matt Low	07c96082e7	Add LastMessageAt field to conversation Replaced `LatestConversationMessages` with `LoadConversationList`, which utilizes `LastMessageAt` for much faster conversation loading in the conversation listing TUI and `lmcli list` command.	2024-10-22 17:53:13 +00:00
Matt Low	0384c7cb66	Large refactor - it compiles! This refactor splits out all conversation concerns into a new `conversation` package. There is now a split between `conversation` and `api`s representation of `Message`, the latter storing the minimum information required for interaction with LLM providers. There is necessary conversation between the two when making LLM calls.	2024-10-22 17:53:13 +00:00
Matt Low	2ea8a73eb5	Final(?) chat view footer fix	2024-10-22 17:53:13 +00:00
Matt Low	c9a7eee090	Update go.mod	2024-10-20 22:39:35 +00:00
Matt Low	ae1e85e166	Update Anthropic token counting logic	2024-10-20 02:38:25 +00:00
Matt Low	304820c919	tui: Show selected message indicator only when focused on messages	2024-10-03 16:47:28 +00:00
Matt Low	93c2fb3d1e	tui: revamp footer (some more) Simplified layout logic, reorganized elements	2024-10-03 16:47:28 +00:00
Matt Low	bb48bc9abd	Display generation model in message header and other tweaks Adjusted `ctrl+t` in chat view to toggle `showDetails` which toggles the display of system messages, message metadata (generation model), and tool call details Modified message selection update logic to skip messages that aren't shown	2024-09-30 20:26:09 +00:00
Matt Low	5d13c3e056	Add metadata json field to Message, store generation model/provider	2024-09-30 17:44:43 +00:00
Matt Low	327a128b2f	Moved api.ChatCompletionProvider, api.Chunk to api/provider	2024-09-30 16:15:42 +00:00
Matt Low	a441866f2f	Configure database logging to file	2024-09-30 16:10:51 +00:00
Matt Low	ce7b07ad95	tui: rename	2024-09-30 04:32:47 +00:00
Matt Low	2fed682969	tui: Chat view footer rewrite Rewrote footer handling to better handle truncation, and use `ActiveModel` to return the (stylized) active model	2024-09-26 18:32:33 +00:00
Matt Low	69cdc0a5aa	tui: Add setting view with support for changing the current model	2024-09-26 18:32:22 +00:00
Matt Low	3ec2675632	tui: Error handling tweak Moved errors to bottom of screen, fix infinite loop by typing errors properly	2024-09-23 04:43:26 +00:00
Matt Low	172bfc57e1	Allow specifying `--agent none` to mean no agent	2024-09-23 03:04:43 +00:00
Matt Low	a46d211e10	Improve TUI system prompt handling + allow setting a default agent	2024-09-23 03:00:03 +00:00
Matt Low	676aa7b004	Refactor TUI rendering handling and general cleanup Improves render handling by moving the responsibility of laying out the whole UI from each view and into the main `tui` model. Our `ViewModel` interface has now diverged from bubbletea's `Model` and introduces individual `Header`, `Content`, and `Footer` methods for rendering those UI elements. Also moved away from using value receivers on our Update and View functions (as is common across Bubbletea) to pointer receivers, which cleaned up some of the weirder aspects of the code (e.g. before we essentially had no choice but to do our rendering in `Update` in order to calculate and update the final height of the main content's `viewport`).	2024-09-23 02:49:08 +00:00
Matt Low	b7c89a4dd1	Update TODO.md	2024-09-21 20:13:47 +00:00
Matt Low	b8e3172ce0	Start new conversations from TUI	2024-09-21 02:47:03 +00:00
Matt Low	a1fdf3f7cd	Deprecation fix	2024-09-21 02:46:51 +00:00
Matt Low	a488ec4fd8	Fixed message loading Root messages weren't being loaded since the refactor, and there was dead code	2024-09-21 02:32:54 +00:00
Matt Low	463ca9ef40	TUI view management and input handling cleanup	2024-09-16 16:18:18 +00:00
Matt Low	24b5cdbbf6	More monior TUI refactor/cleanup `tui/tui.go` is no longer responsible for passing window resize updates to all views, instead we request a new window size message to be sent at the same time we enter the view, allowing the view to catch and handle it. Add `Initialized` to `tui/shared/View` model, now we only call `Init` on a view before entering it for the first time, rather than calling `Init` on all views when the application starts. Renames file, small cleanups	2024-09-16 14:04:08 +00:00
Matt Low	7c0bfefc65	Update deps	2024-09-16 03:49:04 +00:00
Matt Low	443c8096d3	TUI refactor - Clean up, improved startup logic, initial conversation load - Moved converation/message business logic (mostly) into `model/tui`	2024-09-16 00:48:45 +00:00
Matt Low	1570988b98	Fix LatestConversationMessages preload Load the conversation's selected root as well	2024-09-16 00:37:42 +00:00
Matt Low	434fc4672b	Allow custom headers on OpenAI providers (to be added to more later)	2024-08-12 17:14:53 +00:00
Matt Low	fe838f400f	Minor adjustment to seleted message style	2024-07-10 01:21:06 +00:00
Matt Low	e59ce973b6	Add conversation deletion to conversations view	2024-07-08 06:40:52 +00:00
Matt Low	4ef841e945	Add modify_file tool Removed file_insert_lines, file_replace_lines	2024-07-01 08:09:04 +00:00
Matt Low	c68084f8a5	Update TODO	2024-06-24 17:45:18 +00:00
Matt Low	8ca044b6af	Update README.md	2024-06-23 21:42:23 +00:00
Matt Low	6f5cf68208	Update TODO.md	2024-06-23 21:42:23 +00:00
Matt Low	914d9ac0c1	Renamed RequestParameters.ToolBag to Toolbox	2024-06-23 19:10:03 +00:00
Matt Low	8ddac2f820	Introduce "agents" An agent is currently a name given to a system prompt and a set of tools which the agent has access to. This resolves the previous issue of the set of configured tools being available in all contexts, which wasn't always desired. Tools are now only available when an agent is explicitly requested using the `-a/--agent` flag. Agents are expected to be expanded on: the concept of task-specilized agents (e.g. coding), the ability to define a set of files an agent should always have access to for RAG purposes, etc. Other changes: - Removes the "tools" top-level config structure (though this is expected to come back along with the abillity to define custom tools). - Renamed `pkg/agent` to `pkg/agents`	2024-06-23 19:05:30 +00:00
Matt Low	cea5118cac	Only include user and assistant messages when generating titles (again)	2024-06-23 18:55:04 +00:00
Matt Low	a43a91c6ff	Update system prompt handling (again) Add `api.ApplySystemPrompt`, renamed `GetSystemPrompt` to `DefaultSystemPrompt`.	2024-06-23 18:36:51 +00:00
Matt Low	ba7018af11	Update config handling - Stop using pointers where unnecessary - Removed default system prompt - Set indent level to 2 when writing config - Update ordering of config struct, which affects marshalling - Make provider `name` optional, defaulting to the provider's `kind`	2024-06-23 16:28:21 +00:00
Matt Low	f89cc7b410	Add validation to command line flags + update system prompt handling Renamed `applyPromptFlags` to `applyGenerationFlags` and added `validateGenerationFlags`	2024-06-23 06:08:15 +00:00
Matt Low	677cfcfebf	Slight cleanup to openai Remove /v1 from base url, removed some slight repetition	2024-06-23 04:17:53 +00:00
Matt Low	11402c5534	Update to yaml.v3 Bonus: better rendering of tool results in the chat	2024-06-23 04:04:01 +00:00
Matt Low	a1fc8a637b	Include types in provider files, splitting wasn't ergonomic	2024-06-23 01:48:31 +00:00
Matt Low	94d84ba7d7	Support Anthropic's native tool calling API	2024-06-23 01:47:31 +00:00
Matt Low	c50b6b154d	Add TODO.md	2024-06-21 12:56:11 -06:00
Matt Low	31df055430	Always show tool calls, toggle whether results are hidden	2024-06-21 06:05:00 +00:00
Matt Low	c30e652103	Cleaned up assistant cursor handling	2024-06-21 05:52:59 +00:00
Matt Low	3fde58b77d	Package restructure and API changes, several fixes - More emphasis on `api` package. It now holds database model structs from `lmcli/models` (which is now gone) as well as the tool spec, call, and result types. `tools.Tool` is now `api.ToolSpec`. `api.ChatCompletionClient` was renamed to `api.ChatCompletionProvider`. - Change ChatCompletion interface and implementations to no longer do automatic tool call recursion - they simply return a ToolCall message which the caller can decide what to do with (e.g. prompt for user confirmation before executing) - `api.ChatCompletionProvider` functions have had their ReplyCallback parameter removed, as now they only return a single reply. - Added a top-level `agent` package, moved the current built-in tools implementations under `agent/toolbox`. `tools.ExecuteToolCalls` is now `agent.ExecuteToolCalls`. - Fixed request context handling in openai, google, ollama (use `NewRequestWithContext`), cleaned up request cancellation in TUI - Fix tool call tui persistence bug (we were skipping message with empty content) - Now handle tool calling from TUI layer TODO: - Prompt users before executing tool calls - Automatically send tool results to the model (or make this toggleable)	2024-06-21 05:24:02 +00:00
Matt Low	85a2abbbf3	Use @ as the separator between model and provider Also put the provider after the model (e.g. `gpt-4o@openai`, `openai/gpt-4o@openrouter`. The aim here was to reduce clashing and confusion with existing model naming conventions.	2024-06-20 23:37:11 +00:00
Matt Low	dfe43179c0	Include token count in api.Chunk And calculate the tokens/chunk for gemini responses, fixing the tok/s meter for gemini models. Further, only consider the first candidate of streamed gemini responses.	2024-06-09 20:49:18 +00:00
Matt Low	42c3297e54	Make Conversation a pointer refernece on Message Instead of a value, which lead some odd handling of conversation references. Also fixed some formatting and removed an unnecessary (and probably broken) setting of ConversationID in a call to `cmdutil.HandleConversationReply`	2024-06-09 18:51:44 +00:00
Matt Low	a22119f738	Better handling of newly saved conversations When a new conversation is created in the chat view's `persistConversation`, we now set `rootMessages` appropriately.	2024-06-09 18:51:44 +00:00
Matt Low	a2c860252f	Refactor pkg/lmcli/provider Moved `ChangeCompletionInterface` to `pkg/api`, moved individual providers to `pkg/api/provider`	2024-06-09 18:31:43 +00:00
Matt Low	d2d946b776	Wrap chunk content in a Chunk type Preparing to include additional information with each chunk (e.g. token count)	2024-06-09 18:31:43 +00:00
Matt Low	c963747066	Store fixes We were taking double pointers (`**T`) in some areas, and in we were not setting foreign references correctly in `StartConversation` and `Reply`.	2024-06-09 18:31:40 +00:00
Matt Low	e334d9fc4f	Remove forgotten printf	2024-06-09 16:19:22 +00:00
Matt Low	c1ead83939	Rename shared.State to shared.Shared	2024-06-09 16:19:19 +00:00
Matt Low	c9e92e186e	Chat view cleanup Replace `waitingForReply` and the `status` string with the `state` variable.	2024-06-09 16:19:17 +00:00
Matt Low	45df957a06	Fixes to message/conversation handling in tui chat view This set of changes fixes root/child message cycling and ensures all database operations happen within a `tea.Cmd`	2024-06-08 21:28:29 +00:00
Matt Low	136c463924	Split chat view into files	2024-06-02 22:40:46 +00:00
Matt Low	2580087b4d	Fixed gemini system prompt handling	2024-06-02 22:37:50 +00:00
Matt Low	60a474d516	Implement PathToRoot and PathToLeaf with one query After fetching all of a conversation's messages, we traverse the message's Parent or SelectedReply fields to build the message "path" in-memory	2024-06-01 06:40:59 +00:00
Matt Low	ea576d24a6	Add Ollama support	2024-06-01 01:38:45 +00:00
Matt Low	465b1d333e	Fixed handling of long (slash separated) and short model identifiers Renamed `GetCompletionProvider` to `GetModelProvider` and update it to return the model's short name (the one to use when making requests)	2024-05-30 19:06:18 +00:00
Matt Low	b29a4c8b84	Fixed regression from `3536438d` We were sending an empty string to the output channel when `ping` messages were received from Anthropic's API. This was causing the TUI to break since we started doing an empty chunk check (and mistakenly not waiting for future chunks if one was received). This commit makes it so we no longer an empty string on the ping message from Anthropic, and we update the handling of msgAssistantChunk and msgAssistantReply to make it less likely that we forget to wait for the next chunk/reply.	2024-05-30 18:58:03 +00:00
Matt Low	58e1b84fea	Documentation tweak	2024-05-30 18:24:01 +00:00
Matt Low	a6522dbcd0	Generate title prompt tweak	2024-05-30 18:24:01 +00:00
Matt Low	97cd047861	Cleaned up tui view switching	2024-05-30 07:18:31 +00:00
Matt Low	ed784bb1cf	Clean up tui View handling	2024-05-30 07:05:08 +00:00
Matt Low	c1792f27ff	Split up tui code into packages (views/*, shared, util)	2024-05-30 06:44:40 +00:00
Matt Low	0ad698a942	Update GenerateTitle Show conversation and expect result back in JSON	2024-05-28 07:37:09 +00:00
Matt Low	0d66a49997	Add ability to cycle through conversation branches in tui	2024-05-28 06:34:11 +00:00
Matt Low	008fdc0d37	Update title generation prompt	2024-05-23 06:01:30 +00:00
Matt Low	eec9eb41e9	Tiny formatting fix	2024-05-23 05:53:13 +00:00
Matt Low	437997872a	Improve message wrapping behavior	2024-05-22 16:57:52 +00:00
Matt Low	3536438dd1	Add cursor to indicate the assistant is responding A better/more natural indication that the model is doing something	2024-05-22 16:25:16 +00:00
Matt Low	f5ce970102	Set default retry offset to 0	2024-05-21 00:13:56 +00:00
Matt Low	5c1248184b	Update dir_tree to have maximum depth of 5 Until we have some mechanism in place for confirming tool calls with the user before executing, it's dangerous to allow unlimited depth	2024-05-21 00:08:42 +00:00
Matt Low	8c53752146	Add message branching Updated the behaviour of commands: - `lmcli edit` - by default create a new branch/message branch with the edited contents - add --in-place to avoid creating a branch - no longer delete messages after the edited message - only do the edit, don't fetch a new response - `lmcli retry` - create a new branch rather than replacing old messages - add --offset to change where to retry from	2024-05-20 22:29:51 +00:00
Matt Low	f6e55f6bff	`lmcli chat`: check that conversation exists	2024-05-20 16:07:38 +00:00
Matt Low	dc1edf8c3e	Split google API types into types.go	2024-05-19 21:50:43 +00:00
Matt Low	62d98289e8	Fix for non-streamed gemini responses	2024-05-19 02:59:43 +00:00
Matt Low	b82f3019f0	Trim space when generating title	2024-05-19 02:59:16 +00:00
Matt Low	1bd953676d	Add name prefix and / separator (e.g. anthropic/claude-3-haiku...)	2024-05-19 02:39:07 +00:00
Matt Low	a291e7b42c	Gemini cleanup, tool calling working	2024-05-19 01:38:02 +00:00
Matt Low	1b8d04c96d	Gemini fixes, tool calling	2024-05-18 23:18:53 +00:00
Matt Low	cbcd3b1ba9	Gemini WIP	2024-05-18 22:14:41 +00:00
Matt Low	75bf9f6125	Tweaks to read_file and dir_tree	2024-05-14 23:00:00 +00:00
Matt Low	9ff4322995	Formatting	2024-05-14 20:55:11 +00:00
Matt Low	54f5a3c209	Improved util.SetSTructDefaults	2024-05-14 20:54:37 +00:00
Matt Low	86bdc733bf	Add token/sec counter to tui	2024-05-14 03:41:19 +00:00
Matt Low	60394de620	Listen for msgStateEnter in conversations view	2024-05-08 13:32:44 +00:00
Matt Low	aeeb7bb7f7	tui: Add --system-prompt handling And some state handling changes	2024-05-07 08:19:45 +00:00
Matt Low	2b38db7db7	Update command flag handling `lmcli chat` now supports common prompt flags (model, length, system prompt, etc)	2024-05-07 08:18:48 +00:00
Matt Low	8e4ff90ab4	Multiple provider configuration Add support for having multiple openai or anthropic compatible providers accessible via different baseUrls	2024-05-05 08:15:17 +00:00
Matt Low	bdaf6204f6	Add openai response error handling	2024-05-05 07:32:35 +00:00
Matt Low	1b9a8f319c	Split anthropic types out to types.go	2024-04-29 06:16:41 +00:00
Matt Low	ffe9d299ef	Remove go-openai	2024-04-29 06:14:36 +00:00
Matt Low	08a2027332	tui: cleanup	2024-04-03 07:10:41 +00:00
Matt Low	b06e031ee0	tui: Update conversation list category heading colour	2024-04-03 07:06:25 +00:00
Matt Low	69d3265b64	tui: fleshed out converation selection	2024-04-02 07:04:12 +00:00
Matt Low	7463b7502c	tui: basic conversation selection and navigation	2024-04-01 22:47:15 +00:00
Matt Low	0e68e22efa	tui: cleanup conversations data model	2024-04-01 22:43:20 +00:00
Matt Low	1404cae6a7	tui: call handleResize on states before transitioning	2024-04-01 17:07:50 +00:00
Matt Low	9e6d41a3ff	tui: fixed Init handling Don't re-init components on each state change	2024-04-01 17:03:49 +00:00
Matt Low	39cd4227c6	tui: fix wrapping	2024-04-01 16:42:23 +00:00
Matt Low	105ee2e01b	tui: update/clean up input handling	2024-04-01 16:42:23 +00:00
Matt Low	e1970a315a	tui: split model up into chat/conversations	2024-03-31 23:51:45 +00:00
Matt Low	020db40401	tui: renamed stateConversation -> stateChat stateConversationList -> stateConversations	2024-03-30 20:50:33 -06:00
Matt Low	811ec4b251	tui: split up conversation related code into conversation.go moved some things to util, re-ordered some functions	2024-03-30 20:50:33 -06:00
Matt Low	c68cb14eb9	tui: Initial rough conversation list view	2024-03-30 20:50:33 -06:00
Matt Low	cef87a55d8	tui: initial wiring of different "app states"	2024-03-30 20:50:33 -06:00
Matt Low	29519fa2f3	Add -a/-c shorthands for `lmcli list --all/--count`	2024-03-30 20:50:20 -06:00
Matt Low	2e3779ad32	tui: remove temporary edit file	2024-03-29 22:26:28 +00:00
Matt Low	9cd28d28d7	tui: renamed uiCache to views, cleanup	2024-03-29 20:56:39 +00:00
Matt Low	0b991800d6	tui: dynamic input textarea height and styling updates Maintain a height of 4 up to half of the main content area Add rounded border	2024-03-29 20:00:28 +00:00
Matt Low	5af857edae	tui: truncate title to width	2024-03-29 15:48:50 +00:00
Matt Low	3e24a54d0a	tui: add border above input	2024-03-28 06:53:39 +00:00
Matt Low	a669313a0b	tui: add tool rendering cleaned up message rendering and changed cache semantics other smaller tweaks	2024-03-26 08:06:46 +00:00
Matt Low	6310021dca	tui: improve footer truncation	2024-03-23 04:08:48 +00:00
Matt Low	ef929da68c	tui: add uiCache Clean up/fix how we calculate the height of the content viewport	2024-03-23 03:55:20 +00:00
Matt Low	c51644e78e	Add dir_tree tool	2024-03-22 20:30:34 +00:00
Matt Low	91c74d9e1e	Update CreateChatCompletion behavior When the last message in the passed messages slice is an assistant message, treat it as a partial message that is being continued, and include its content in the newly created reply Update TUI code to handle new behavior	2024-03-22 20:02:28 +00:00