tell-me/llm/client.go

package llm

import (
	"context"
	"encoding/json"
	"fmt"
	"io"
	"strings"
	"time"

	"tell-me/mcp"
	"tell-me/tools"

	"github.com/sashabaranov/go-openai"
)

// Client wraps the OpenAI client for LLM interactions
type Client struct {
	client      *openai.Client
	model       string
	contextSize int
	searxngURL  string
	mcpManager  *mcp.Manager
}

// NewClient creates a new LLM client
func NewClient(apiURL, apiKey, model string, contextSize int, searxngURL string, mcpManager *mcp.Manager) *Client {
	config := openai.DefaultConfig(apiKey)
	config.BaseURL = apiURL

	client := openai.NewClientWithConfig(config)

	return &Client{
		client:      client,
		model:       model,
		contextSize: contextSize,
		searxngURL:  searxngURL,
		mcpManager:  mcpManager,
	}
}

// GetSystemPrompt returns the system prompt with current date appended
func GetSystemPrompt(prompt string) string {
	currentDate := time.Now().Format("2006-01-02")
	return fmt.Sprintf("%s\n\nCURRENT DATE: %s", prompt, currentDate)
}

// GetTools returns the tool definitions for the LLM (built-in tools only)
func GetBuiltInTools() []openai.Tool {
	return []openai.Tool{
		{
			Type: openai.ToolTypeFunction,
			Function: &openai.FunctionDefinition{
				Name:        "web_search",
				Description: "Search the internet for information using SearXNG. Use this tool to find current information, facts, news, or any knowledge you need to answer the user's question.",
				Parameters: json.RawMessage(`{
					"type": "object",
					"properties": {
						"query": {
							"type": "string",
							"description": "The search query to find relevant information"
						}
					},
					"required": ["query"]
				}`),
			},
		},
		{
			Type: openai.ToolTypeFunction,
			Function: &openai.FunctionDefinition{
				Name:        "fetch_articles",
				Description: "Fetch and read content from 1-5 articles at once. Provide both titles and URLs from search results. The HTML will be converted to clean text format and combined. Use this after searching to read the most relevant pages together.",
				Parameters: json.RawMessage(`{
					"type": "object",
					"properties": {
						"articles": {
							"type": "array",
							"items": {
								"type": "object",
								"properties": {
									"title": {
										"type": "string",
										"description": "The title of the article from search results"
									},
									"url": {
										"type": "string",
										"description": "The URL to fetch (must start with http:// or https://)"
									}
								},
								"required": ["title", "url"]
							},
							"description": "Array of articles with titles and URLs (1-5 recommended, max 5)"
						}
					},
					"required": ["articles"]
				}`),
			},
		},
	}
}

// GetTools returns all available tools (built-in + MCP tools)
func (c *Client) GetTools() []openai.Tool {
	tools := GetBuiltInTools()

	// Add MCP tools if manager is available
	if c.mcpManager != nil {
		mcpTools := c.mcpManager.GetAllTools()
		tools = append(tools, mcpTools...)
	}

	return tools
}

// Chat sends a message and streams the response, handling tool calls
func (c *Client) Chat(ctx context.Context, messages []openai.ChatCompletionMessage, streamFunc func(string)) (string, []openai.ChatCompletionMessage, error) {
	req := openai.ChatCompletionRequest{
		Model:    c.model,
		Messages: messages,
		Tools:    c.GetTools(),
		Stream:   true,
	}

	stream, err := c.client.CreateChatCompletionStream(ctx, req)
	if err != nil {
		return "", messages, fmt.Errorf("chat completion stream failed: %w", err)
	}
	defer stream.Close()

	var fullContent strings.Builder
	var toolCalls []openai.ToolCall

	for {
		response, err := stream.Recv()
		if err == io.EOF {
			break
		}
		if err != nil {
			return "", messages, fmt.Errorf("stream error: %w", err)
		}

		if len(response.Choices) == 0 {
			continue
		}

		delta := response.Choices[0].Delta

		// Handle content streaming
		if delta.Content != "" {
			fullContent.WriteString(delta.Content)
			if streamFunc != nil {
				streamFunc(delta.Content)
			}
		}

		// Handle tool calls
		if len(delta.ToolCalls) > 0 {
			for _, tc := range delta.ToolCalls {
				if tc.Index != nil {
					idx := *tc.Index

					// Ensure we have enough space in the slice
					for len(toolCalls) <= idx {
						toolCalls = append(toolCalls, openai.ToolCall{})
					}

					if tc.ID != "" {
						toolCalls[idx].ID = tc.ID
					}
					if tc.Type != "" {
						toolCalls[idx].Type = tc.Type
					}
					if tc.Function.Name != "" {
						toolCalls[idx].Function.Name = tc.Function.Name
					}
					if tc.Function.Arguments != "" {
						toolCalls[idx].Function.Arguments += tc.Function.Arguments
					}
				}
			}
		}
	}

	// Create the assistant message
	assistantMsg := openai.ChatCompletionMessage{
		Role:    openai.ChatMessageRoleAssistant,
		Content: fullContent.String(),
	}

	if len(toolCalls) > 0 {
		assistantMsg.ToolCalls = toolCalls
	}

	messages = append(messages, assistantMsg)

	// Handle tool calls if present
	if len(toolCalls) > 0 {
		for _, toolCall := range toolCalls {
			result := c.handleToolCall(ctx, toolCall)

			// Add tool response to messages
			messages = append(messages, openai.ChatCompletionMessage{
				Role:       openai.ChatMessageRoleTool,
				Content:    result,
				ToolCallID: toolCall.ID,
			})
		}

		// Print blank line before streaming the final response
		fmt.Println()

		// Make another streaming call with tool results
		return c.Chat(ctx, messages, streamFunc)
	}

	return fullContent.String(), messages, nil
}

// handleToolCall routes tool calls to the appropriate handler
func (c *Client) handleToolCall(ctx context.Context, toolCall openai.ToolCall) string {
	toolName := toolCall.Function.Name

	// Check if it's a built-in tool
	switch toolName {
	case "web_search":
		var args struct {
			Query string `json:"query"`
		}
		if err := json.Unmarshal([]byte(toolCall.Function.Arguments), &args); err != nil {
			return fmt.Sprintf("Error parsing arguments: %v", err)
		}
		fmt.Printf("Searching: %s\n", args.Query)
		result, err := tools.WebSearch(c.searxngURL, args.Query)
		if err != nil {
			return fmt.Sprintf("Search error: %v", err)
		}
		return result

	case "fetch_articles":
		var args struct {
			Articles []struct {
				Title string `json:"title"`
				URL   string `json:"url"`
			} `json:"articles"`
		}
		if err := json.Unmarshal([]byte(toolCall.Function.Arguments), &args); err != nil {
			return fmt.Sprintf("Error parsing arguments: %v", err)
		}
		fmt.Printf("Reading %d articles:\n", len(args.Articles))
		urls := make([]string, len(args.Articles))
		for i, article := range args.Articles {
			fmt.Printf("    - %s\n", article.Title)
			urls[i] = article.URL
		}
		result, err := tools.FetchArticles(urls)
		if err != nil {
			return fmt.Sprintf("Fetch error: %v", err)
		}
		return result

	default:
		// Check if it's an MCP tool (format: servername_toolname)
		if c.mcpManager != nil && strings.Contains(toolName, "_") {
			fmt.Printf("Calling MCP tool: %s\n", toolName)
			result, err := c.mcpManager.CallTool(ctx, toolName, toolCall.Function.Arguments)
			if err != nil {
				return fmt.Sprintf("MCP tool error: %v", err)
			}
			return result
		}

		return fmt.Sprintf("Unknown tool: %s", toolName)
	}
}