Instead of CreateChatCompletion* accepting a pointer to a slice of reply
messages, it accepts a callback which is called with each successive
reply the conversation.
This gives the caller more flexibility in how it handles replies (e.g.
it can react to them immediately now, instead of waiting for the entire
call to finish)
- Split pkg/cli/cmd.go into new pkg/cmd package
- Split pkg/cli/functions.go into pkg/lmcli/tools package
- Refactor pkg/cli/openai.go to pkg/lmcli/provider/openai
Other changes:
- Made models configurable
- Slight config reorganization
Allows replies (tool calls, user-facing messges) to be added in sequence
as CreateChatCompleion* recurses into itself.
Cleaned up cmd.go: no longer need to create a Message based on the
string content response.
Adds the following tools:
- read_dir - list a directory's contents
- read_file - read the content of a file
- write_file - write contents to a file
- insert_file_lines - insert lines in a file
- replace_file_lines - replace or remove lines in a file
Update CreateChangeCompletionStream to return the entire response upon
stream completion. Renamed HandleDelayedResponse to
HandleDelayedContent, which no longer returns the content.
Removes the need wrapping HandleDelayedContent in an immediately invoked
function and the passing of the completed response over a channel. Also
allows us to better handle the case of partial a response.
These allow to set a different system prompt for conversations and
one-shot prompts.
Also add a new `modelDefaults.systemPrompt` configuration key to define
the default system prompt, which can be overriden per-execution with the
--system-prompt or --system-prompt-file flags.
Add new SetStructDefaults function to handle the "defaults" struct tag.
Only works on struct fields which are pointers (in order to be able to
distinguish between not set (nil) and zero values). So, the Config
struct has been updated to use pointer fields and we now need to
dereference those pointers to use them.
