vetrag/llm.go

package main

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
	"strings"
	"text/template"

	"github.com/sirupsen/logrus"
)

// LLMClient abstracts LLM API calls
type LLMClient struct {
	APIKey  string
	BaseURL string
	Model   string
}

// NewLLMClient constructs a new LLMClient with the given API key and base URL
func NewLLMClient(apiKey, baseURL string, model string) *LLMClient {
	return &LLMClient{
		APIKey:  apiKey,
		BaseURL: baseURL,
		Model:   model,
	}
}

// renderPrompt renders a Go template with the given data
func renderPrompt(tmplStr string, data any) (string, error) {
	tmpl, err := template.New("").Parse(tmplStr)
	if err != nil {
		return "", err
	}
	var buf bytes.Buffer
	if err := tmpl.Execute(&buf, data); err != nil {
		return "", err
	}
	return buf.String(), nil
}

// ExtractKeywords calls LLM to extract keywords from user message
func (llm *LLMClient) ExtractKeywords(ctx context.Context, message string) (map[string]interface{}, error) {
	prompt, err := renderPrompt(appConfig.LLM.ExtractKeywordsPrompt, map[string]string{"Message": message})
	if err != nil {
		logrus.WithError(err).Error("[CONFIG] Failed to render ExtractKeywords prompt")
		return nil, err
	}
	logrus.WithField("prompt", prompt).Info("[LLM] ExtractKeywords prompt")
	format := map[string]interface{}{
		"type": "object",
		"properties": map[string]interface{}{
			"translate": map[string]interface{}{"type": "string"},
			"keyword":   map[string]interface{}{"type": "array", "items": map[string]interface{}{"type": "string"}},
			"animal":    map[string]interface{}{"type": "string"},
		},
		"required": []string{"translate", "keyword", "animal"},
	}
	resp, err := llm.openAICompletion(ctx, prompt, format)
	logrus.WithFields(logrus.Fields{"response": resp, "err": err}).Info("[LLM] ExtractKeywords response")
	if err != nil {
		return nil, err
	}
	var result map[string]interface{}
	if err := json.Unmarshal([]byte(resp), &result); err != nil {
		return nil, err
	}
	return result, nil
}

// DisambiguateBestMatch calls LLM to pick best match from candidates
func (llm *LLMClient) DisambiguateBestMatch(ctx context.Context, message string, candidates []Visit) (string, error) {
	format := map[string]interface{}{
		"type": "object",
		"properties": map[string]interface{}{
			"visitReason": map[string]interface{}{"type": "string"},
		},
		"required": []string{"visitReason"},
	}
	entries, _ := json.Marshal(candidates)
	prompt, err := renderPrompt(appConfig.LLM.DisambiguatePrompt, map[string]string{"Entries": string(entries), "Message": message})
	if err != nil {
		logrus.WithError(err).Error("[CONFIG] Failed to render Disambiguate prompt")
		return "", err
	}
	logrus.WithField("prompt", prompt).Info("[LLM] DisambiguateBestMatch prompt")
	resp, err := llm.openAICompletion(ctx, prompt, format)
	logrus.WithFields(logrus.Fields{"response": resp, "err": err}).Info("[LLM] DisambiguateBestMatch response")
	if err != nil {
		return "", err
	}
	var parsed map[string]string
	if err := json.Unmarshal([]byte(resp), &parsed); err != nil {
		return "", fmt.Errorf("failed to unmarshal disambiguation response: %w", err)
	}

	visitReason := strings.TrimSpace(parsed["visitReason"])
	if visitReason == "" {
		return "", fmt.Errorf("visitReason not found in response")
	}

	return visitReason, nil
}

// openAICompletion now supports both Ollama (default local) and OpenRouter/OpenAI-compatible APIs without external branching.
// It auto-detects by inspecting the BaseURL. If the URL contains "openrouter.ai" or "/v1/", it assumes OpenAI-style.
func (llm *LLMClient) openAICompletion(ctx context.Context, prompt string, format map[string]interface{}) (string, error) {
	apiURL := llm.BaseURL
	if apiURL == "" {
		// Default to Ollama local chat endpoint
		apiURL = "http://localhost:11434/api/chat"
	}

	isOpenAIStyle := strings.Contains(apiURL, "openrouter.ai") || strings.Contains(apiURL, "/v1/")

	// Build request body depending on style
	var body map[string]interface{}
	if isOpenAIStyle {
		// OpenAI / OpenRouter style (chat.completions)
		// Use response_format with JSON schema when provided.
		responseFormat := map[string]interface{}{
			"type": "json_schema",
			"json_schema": map[string]interface{}{
				"name":   "structured_output",
				"schema": format,
			},
		}
		body = map[string]interface{}{
			"model":           llm.Model,
			"messages":        []map[string]string{{"role": "user", "content": prompt}},
			"response_format": responseFormat,
		}
	} else {
		// Ollama structured output extension
		body = map[string]interface{}{
			"model":    llm.Model,
			"messages": []map[string]string{{"role": "user", "content": prompt}},
			"stream":   false,
			"format":   format,
		}
	}

	jsonBody, _ := json.Marshal(body)
	logrus.WithFields(logrus.Fields{"api_url": apiURL, "prompt": prompt, "is_openai_style": isOpenAIStyle}).Info("[LLM] completion POST")

	req, _ := http.NewRequestWithContext(ctx, http.MethodPost, apiURL, bytes.NewBuffer(jsonBody))
	if llm.APIKey != "" {
		// OpenRouter expects: Authorization: Bearer sk-... or OR-... depending on key type
		req.Header.Set("Authorization", "Bearer "+llm.APIKey)
	}
	req.Header.Set("Content-Type", "application/json")

	client := &http.Client{}
	resp, err := client.Do(req)
	if err != nil {
		logrus.WithError(err).Error("[LLM] completion HTTP error")
		return "", err
	}
	defer resp.Body.Close()

	raw, err := io.ReadAll(resp.Body)
	if err != nil {
		return "", fmt.Errorf("failed reading response body: %w", err)
	}
	logrus.WithFields(logrus.Fields{"status": resp.StatusCode, "raw": string(raw)}).Debug("[LLM] completion raw response")

	// Attempt Ollama format first (backwards compatible)
	var ollama struct {
		Message struct {
			Content string `json:"content"`
		} `json:"message"`
	}
	if err := json.Unmarshal(raw, &ollama); err == nil && ollama.Message.Content != "" {
		logrus.WithField("content", ollama.Message.Content).Info("[LLM] completion (ollama) parsed")
		return ollama.Message.Content, nil
	}

	// Attempt OpenAI / OpenRouter style
	var openAI struct {
		Choices []struct {
			Message struct {
				Role    string `json:"role"`
				Content string `json:"content"`
			} `json:"message"`
		} `json:"choices"`
		Error *struct {
			Message string `json:"message"`
			Type    string `json:"type"`
		} `json:"error"`
	}
	if err := json.Unmarshal(raw, &openAI); err == nil {
		if openAI.Error != nil {
			return "", fmt.Errorf("provider error: %s (%s)", openAI.Error.Message, openAI.Error.Type)
		}
		if len(openAI.Choices) > 0 && openAI.Choices[0].Message.Content != "" {
			content := openAI.Choices[0].Message.Content
			logrus.WithField("content", content).Info("[LLM] completion (openai) parsed")
			return content, nil
		}
	}

	// If still nothing, return error with snippet
	return "", fmt.Errorf("unrecognized LLM response format: %.200s", string(raw))
}

// LLMClientAPI allows mocking LLMClient in other places
// Only public methods should be included

type LLMClientAPI interface {
	ExtractKeywords(ctx context.Context, message string) (map[string]interface{}, error)
	DisambiguateBestMatch(ctx context.Context, message string, candidates []Visit) (string, error)
}

var _ LLMClientAPI = (*LLMClient)(nil)