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)
2024-06-12 08:35:07 +00:00
parent 85a2abbbf3
commit 3fde58b77d
35 changed files with 608 additions and 749 deletions
--- 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"
 )

@@ -36,7 +36,7 @@ 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")
 			}

@@ -50,7 +50,7 @@ func ContinueCmd(ctx *lmcli.Context) *cobra.Command {
 			}

 			// 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)
--- 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"
 )

@@ -53,10 +53,10 @@ func EditCmd(ctx *lmcli.Context) *cobra.Command {

 			role, _ := cmd.Flags().GetString("role")
 			if role != "" {
-				if role != string(model.MessageRoleUser) && role != string(model.MessageRoleAssistant) {
+				if role != string(api.MessageRoleUser) && role != string(api.MessageRoleAssistant) {
 					return fmt.Errorf("Invalid role specified. Please use 'user' or 'assistant'.")
 				}
-				toEdit.Role = model.MessageRole(role)
+				toEdit.Role = api.MessageRole(role)
 			}

 			// Update the message in-place
--- a/pkg/cmd/new.go
+++ b/pkg/cmd/new.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"
 )

@@ -20,19 +20,19 @@ func NewCmd(ctx *lmcli.Context) *cobra.Command {
 				return fmt.Errorf("No message was provided.")
 			}

-			var messages []model.Message
+			var messages []api.Message

 			// TODO: probably just make this part of the conversation
 			system := ctx.GetSystemPrompt()
 			if system != "" {
-				messages = append(messages, model.Message{
-					Role:    model.MessageRoleSystem,
+				messages = append(messages, api.Message{
+					Role:    api.MessageRoleSystem,
 					Content: system,
 				})
 			}

-			messages = append(messages, model.Message{
-				Role:    model.MessageRoleUser,
+			messages = append(messages, api.Message{
+				Role:    api.MessageRoleUser,
 				Content: input,
 			})

--- a/pkg/cmd/prompt.go
+++ b/pkg/cmd/prompt.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"
 )

@@ -20,19 +20,19 @@ func PromptCmd(ctx *lmcli.Context) *cobra.Command {
 				return fmt.Errorf("No message was provided.")
 			}

-			var messages []model.Message
+			var messages []api.Message

 			// TODO: stop supplying system prompt as a message
 			system := ctx.GetSystemPrompt()
 			if system != "" {
-				messages = append(messages, model.Message{
-					Role:    model.MessageRoleSystem,
+				messages = append(messages, api.Message{
+					Role:    api.MessageRoleSystem,
 					Content: system,
 				})
 			}

-			messages = append(messages, model.Message{
-				Role:    model.MessageRoleUser,
+			messages = append(messages, api.Message{
+				Role:    api.MessageRoleUser,
 				Content: input,
 			})

--- a/pkg/cmd/remove.go
+++ b/pkg/cmd/remove.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"
 )

@@ -23,7 +23,7 @@ func RemoveCmd(ctx *lmcli.Context) *cobra.Command {
 			return nil
 		},
 		RunE: func(cmd *cobra.Command, args []string) error {
-			var toRemove []*model.Conversation
+			var toRemove []*api.Conversation
 			for _, shortName := range args {
 				conversation := cmdutil.LookupConversation(ctx, shortName)
 				toRemove = append(toRemove, conversation)
--- a/pkg/cmd/reply.go
+++ b/pkg/cmd/reply.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"
 )

@@ -30,8 +30,8 @@ func ReplyCmd(ctx *lmcli.Context) *cobra.Command {
 				return fmt.Errorf("No reply was provided.")
 			}

-			cmdutil.HandleConversationReply(ctx, conversation, true, model.Message{
-				Role:    model.MessageRoleUser,
+			cmdutil.HandleConversationReply(ctx, conversation, true, api.Message{
+				Role:    api.MessageRoleUser,
 				Content: reply,
 			})
 			return nil
--- 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"
 )

@@ -43,11 +43,11 @@ func RetryCmd(ctx *lmcli.Context) *cobra.Command {
 			retryFromIdx := len(messages) - 1 - offset

 			// decrease retryFromIdx until we hit a user message
-			for retryFromIdx >= 0 && messages[retryFromIdx].Role != model.MessageRoleUser {
+			for retryFromIdx >= 0 && messages[retryFromIdx].Role != api.MessageRoleUser {
 				retryFromIdx--
 			}

-			if messages[retryFromIdx].Role != model.MessageRoleUser {
+			if messages[retryFromIdx].Role != api.MessageRoleUser {
 				return fmt.Errorf("No user messages to retry")
 			}

--- a/pkg/cmd/util/util.go
+++ b/pkg/cmd/util/util.go
@@ -10,36 +10,36 @@ import (

 	"git.mlow.ca/mlow/lmcli/pkg/api"
 	"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"
 )

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

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

-	response, err := provider.CreateChatCompletionStream(
-		context.Background(), requestParams, messages, callback, content,
+	content := make(chan api.Chunk)
+	defer close(content)
+
+	// render the content received over the channel
+	go ShowDelayedContent(content)
+
+	reply, err := provider.CreateChatCompletionStream(
+		context.Background(), requestParams, messages, content,
 	)
-	if response != "" {
+
+	if reply.Content != "" {
 		// there was some content, so break to a new line after it
 		fmt.Println()

@@ -48,12 +48,12 @@ func Prompt(ctx *lmcli.Context, messages []model.Message, callback func(model.Me
 			err = nil
 		}
 	}
-	return response, err
+	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 {
+func LookupConversation(ctx *lmcli.Context, shortName string) *api.Conversation {
 	c, err := ctx.Store.ConversationByShortName(shortName)
 	if err != nil {
 		lmcli.Fatal("Could not lookup conversation: %v\n", err)
@@ -64,7 +64,7 @@ func LookupConversation(ctx *lmcli.Context, shortName string) *model.Conversatio
 	return c
 }

-func LookupConversationE(ctx *lmcli.Context, shortName string) (*model.Conversation, error) {
+func LookupConversationE(ctx *lmcli.Context, shortName string) (*api.Conversation, error) {
 	c, err := ctx.Store.ConversationByShortName(shortName)
 	if err != nil {
 		return nil, fmt.Errorf("Could not lookup conversation: %v", err)
@@ -75,7 +75,7 @@ func LookupConversationE(ctx *lmcli.Context, shortName string) (*model.Conversat
 	return c, nil
 }

-func HandleConversationReply(ctx *lmcli.Context, c *model.Conversation, persist bool, toSend ...model.Message) {
+func HandleConversationReply(ctx *lmcli.Context, c *api.Conversation, persist bool, toSend ...api.Message) {
 	messages, err := ctx.Store.PathToLeaf(c.SelectedRoot)
 	if err != nil {
 		lmcli.Fatal("Could not load messages: %v\n", err)
@@ -85,7 +85,7 @@ func HandleConversationReply(ctx *lmcli.Context, c *model.Conversation, persist

 // handleConversationReply handles sending messages to an existing
 // conversation, optionally persisting both the sent replies and responses.
-func HandleReply(ctx *lmcli.Context, to *model.Message, persist bool, messages ...model.Message) {
+func HandleReply(ctx *lmcli.Context, to *api.Message, persist bool, messages ...api.Message) {
 	if to == nil {
 		lmcli.Fatal("Can't prompt from an empty message.")
 	}
@@ -97,7 +97,7 @@ func HandleReply(ctx *lmcli.Context, to *model.Message, persist bool, messages .

 	RenderConversation(ctx, append(existing, messages...), true)

-	var savedReplies []model.Message
+	var savedReplies []api.Message
 	if persist && len(messages) > 0 {
 		savedReplies, err = ctx.Store.Reply(to, messages...)
 		if err != nil {
@@ -106,15 +106,15 @@ func HandleReply(ctx *lmcli.Context, to *model.Message, persist bool, messages .
 	}

 	// render a message header with no contents
-	RenderMessage(ctx, (&model.Message{Role: model.MessageRoleAssistant}))
+	RenderMessage(ctx, (&api.Message{Role: api.MessageRoleAssistant}))

-	var lastSavedMessage *model.Message
+	var lastSavedMessage *api.Message
 	lastSavedMessage = to
 	if len(savedReplies) > 0 {
 		lastSavedMessage = &savedReplies[len(savedReplies)-1]
 	}

-	replyCallback := func(reply model.Message) {
+	replyCallback := func(reply api.Message) {
 		if !persist {
 			return
 		}
@@ -131,16 +131,16 @@ func HandleReply(ctx *lmcli.Context, to *model.Message, persist bool, messages .
 	}
 }

-func FormatForExternalPrompt(messages []model.Message, system bool) string {
+func FormatForExternalPrompt(messages []api.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
@@ -150,7 +150,7 @@ func FormatForExternalPrompt(messages []model.Message, system bool) string {
 	return sb.String()
 }

-func GenerateTitle(ctx *lmcli.Context, messages []model.Message) (string, error) {
+func GenerateTitle(ctx *lmcli.Context, messages []api.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:
@@ -177,28 +177,32 @@ Example response:
 		return "", err
 	}

-	generateRequest := []model.Message{
+	generateRequest := []api.Message{
 		{
-			Role:    model.MessageRoleSystem,
+			Role:    api.MessageRoleSystem,
 			Content: systemPrompt,
 		},
 		{
-			Role:    model.MessageRoleUser,
+			Role:    api.MessageRoleUser,
 			Content: string(conversation),
 		},
 	}

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

-	requestParams := model.RequestParameters{
+	requestParams := api.RequestParameters{
 		Model:     m,
 		MaxTokens: 25,
 	}

-	response, err := provider.CreateChatCompletion(context.Background(), requestParams, generateRequest, nil)
+	response, err := provider.CreateChatCompletion(
+		context.Background(), requestParams, generateRequest,
+	)
 	if err != nil {
 		return "", err
 	}
@@ -207,7 +211,7 @@ Example response:
 	var jsonResponse struct {
 		Title string `json:"title"`
 	}
-	err = json.Unmarshal([]byte(response), &jsonResponse)
+	err = json.Unmarshal([]byte(response.Content), &jsonResponse)
 	if err != nil {
 		return "", err
 	}
@@ -272,7 +276,7 @@ func ShowDelayedContent(content <-chan api.Chunk) {
 // 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 []api.Message, spaceForResponse bool) {
 	l := len(messages)
 	for i, message := range messages {
 		RenderMessage(ctx, &message)
@@ -283,7 +287,7 @@ func RenderConversation(ctx *lmcli.Context, messages []model.Message, spaceForRe
 	}
 }

-func RenderMessage(ctx *lmcli.Context, m *model.Message) {
+func RenderMessage(ctx *lmcli.Context, m *api.Message) {
 	var messageAge string
 	if m.CreatedAt.IsZero() {
 		messageAge = "now"
@@ -295,11 +299,11 @@ 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
@@ -20,7 +20,7 @@ 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)