gf-lt/bot.go

package main

import (
	"bufio"
	"bytes"
	"elefant/config"
	"elefant/models"
	"elefant/rag"
	"elefant/storage"
	"encoding/json"
	"fmt"
	"io"
	"log/slog"
	"net/http"
	"os"
	"strings"
	"time"

	"github.com/neurosnap/sentences/english"
	"github.com/rivo/tview"
)

var httpClient = http.Client{}

var (
	cfg                 *config.Config
	logger              *slog.Logger
	activeChatName      string
	chunkChan           = make(chan string, 10)
	streamDone          = make(chan bool, 1)
	chatBody            *models.ChatBody
	store               storage.FullRepo
	defaultFirstMsg     = "Hello! What can I do for you?"
	defaultStarter      = []models.RoleMsg{}
	defaultStarterBytes = []byte{}
	interruptResp       = false
	ragger              *rag.RAG
)

// ====

func formMsg(chatBody *models.ChatBody, newMsg, role string) io.Reader {
	if newMsg != "" { // otherwise let the bot continue
		newMsg := models.RoleMsg{Role: role, Content: newMsg}
		chatBody.Messages = append(chatBody.Messages, newMsg)
		// if rag
		if cfg.RAGEnabled {
			ragResp, err := chatRagUse(newMsg.Content)
			if err != nil {
				logger.Error("failed to form a rag msg", "error", err)
				return nil
			}
			ragMsg := models.RoleMsg{Role: cfg.ToolRole, Content: ragResp}
			chatBody.Messages = append(chatBody.Messages, ragMsg)
		}
	}
	data, err := json.Marshal(chatBody)
	if err != nil {
		logger.Error("failed to form a msg", "error", err)
		return nil
	}
	return bytes.NewReader(data)
}

// func sendMsgToLLM(body io.Reader) (*models.LLMRespChunk, error) {
func sendMsgToLLM(body io.Reader) {
	// nolint
	resp, err := httpClient.Post(cfg.APIURL, "application/json", body)
	if err != nil {
		logger.Error("llamacpp api", "error", err)
		streamDone <- true
		return
	}
	defer resp.Body.Close()
	// llmResp := []models.LLMRespChunk{}
	reader := bufio.NewReader(resp.Body)
	counter := uint32(0)
	for {
		counter++
		if interruptResp {
			interruptResp = false
			logger.Info("interrupted bot response")
			break
		}
		if cfg.ChunkLimit > 0 && counter > cfg.ChunkLimit {
			logger.Warn("response hit chunk limit", "limit", cfg.ChunkLimit)
			streamDone <- true
			break
		}
		llmchunk := models.LLMRespChunk{}
		line, err := reader.ReadBytes('\n')
		if err != nil {
			logger.Error("error reading response body", "error", err)
			continue
		}
		if len(line) <= 1 {
			continue // skip \n
		}
		// starts with -> data:
		line = line[6:]
		if err := json.Unmarshal(line, &llmchunk); err != nil {
			logger.Error("failed to decode", "error", err, "line", string(line))
			streamDone <- true
			return
		}
		// llmResp = append(llmResp, llmchunk)
		// logger.Info("streamview", "chunk", llmchunk)
		// if llmchunk.Choices[len(llmchunk.Choices)-1].FinishReason != "chat.completion.chunk" {
		if llmchunk.Choices[len(llmchunk.Choices)-1].FinishReason == "stop" {
			if llmchunk.Choices[len(llmchunk.Choices)-1].Delta.Content != "" {
				logger.Warn("text inside of finish llmchunk", "chunk", llmchunk, "counter", counter)
			}
			streamDone <- true
			// last chunk
			break
		}
		// bot sends way too many \n
		answerText := strings.ReplaceAll(llmchunk.Choices[0].Delta.Content, "\n\n", "\n")
		chunkChan <- answerText
	}
}

func chatRagUse(qText string) (string, error) {
	tokenizer, err := english.NewSentenceTokenizer(nil)
	if err != nil {
		return "", err
	}
	// TODO: this where llm should find the questions in text and ask them
	questionsS := tokenizer.Tokenize(qText)
	questions := make([]string, len(questionsS))
	for i, q := range questionsS {
		questions[i] = q.Text
	}
	respVecs := []models.VectorRow{}
	for i, q := range questions {
		emb, err := ragger.LineToVector(q)
		if err != nil {
			logger.Error("failed to get embs", "error", err, "index", i, "question", q)
			continue
		}
		// e := &models.EmbeddingResp{
		// 	Embedding: emb,
		// }
		// vecs, err := ragger.SearchEmb(e)
		vecs, err := store.SearchClosest(emb)
		if err != nil {
			logger.Error("failed to query embs", "error", err, "index", i, "question", q)
			continue
		}
		respVecs = append(respVecs, vecs...)
		// logger.Info("returned vector from query search", "question", q, "vec", vec)
	}
	// get raw text
	resps := []string{}
	logger.Info("sqlvec resp", "vecs", respVecs)
	for _, rv := range respVecs {
		resps = append(resps, rv.RawText)
	}
	if len(resps) == 0 {
		return "No related results from vector storage.", nil
	}
	return strings.Join(resps, "\n"), nil
}

func chatRound(userMsg, role string, tv *tview.TextView, regen bool) {
	botRespMode = true
	reader := formMsg(chatBody, userMsg, role)
	if reader == nil {
		logger.Error("empty reader from msgs", "role", role)
		return
	}
	go sendMsgToLLM(reader)
	// if userMsg != "" && !regen { // no need to write assistant icon since we continue old message
	if userMsg != "" || regen {
		fmt.Fprintf(tv, "(%d) ", len(chatBody.Messages))
		fmt.Fprint(tv, cfg.AssistantIcon)
		fmt.Fprint(tv, "\n")
	}
	respText := strings.Builder{}
out:
	for {
		select {
		case chunk := <-chunkChan:
			// fmt.Printf(chunk)
			fmt.Fprint(tv, chunk)
			respText.WriteString(chunk)
			tv.ScrollToEnd()
		case <-streamDone:
			botRespMode = false
			break out
		}
	}
	botRespMode = false
	chatBody.Messages = append(chatBody.Messages, models.RoleMsg{
		Role: cfg.AssistantRole, Content: respText.String(),
	})
	colorText()
	updateStatusLine()
	// bot msg is done;
	// now check it for func call
	// logChat(activeChatName, chatBody.Messages)
	err := updateStorageChat(activeChatName, chatBody.Messages)
	if err != nil {
		logger.Warn("failed to update storage", "error", err, "name", activeChatName)
	}
	findCall(respText.String(), tv)
}

func findCall(msg string, tv *tview.TextView) {
	fc := models.FuncCall{}
	jsStr := toolCallRE.FindString(msg)
	if jsStr == "" {
		return
	}
	prefix := "__tool_call__\n"
	suffix := "\n__tool_call__"
	jsStr = strings.TrimSuffix(strings.TrimPrefix(jsStr, prefix), suffix)
	if err := json.Unmarshal([]byte(jsStr), &fc); err != nil {
		logger.Error("failed to unmarshal tool call", "error", err, "json_string", jsStr)
		return
	}
	// call a func
	f, ok := fnMap[fc.Name]
	if !ok {
		m := fc.Name + "%s is not implemented"
		chatRound(m, cfg.ToolRole, tv, false)
		return
	}
	resp := f(fc.Args...)
	toolMsg := fmt.Sprintf("tool response: %+v", string(resp))
	chatRound(toolMsg, cfg.ToolRole, tv, false)
}

func chatToTextSlice(showSys bool) []string {
	resp := make([]string, len(chatBody.Messages))
	for i, msg := range chatBody.Messages {
		if !showSys && (msg.Role != cfg.AssistantRole && msg.Role != cfg.UserRole) {
			continue
		}
		resp[i] = msg.ToText(i, cfg)
	}
	return resp
}

func chatToText(showSys bool) string {
	s := chatToTextSlice(showSys)
	return strings.Join(s, "")
}

func applyCharCard(cc *models.CharCard) {
	cfg.AssistantRole = cc.Role
	// TODO: need map role->icon
	cfg.AssistantIcon = "<" + cc.Role + ">: "
	// try to load last active chat
	history, err := loadAgentsLastChat(cfg.AssistantRole)
	if err != nil {
		logger.Warn("failed to load last agent chat;", "agent", cc.Role, "err", err)
		history = []models.RoleMsg{
			{Role: "system", Content: cc.SysPrompt},
			{Role: cfg.AssistantRole, Content: cc.FirstMsg},
		}
		id, err := store.ChatGetMaxID()
		if err != nil {
			logger.Error("failed to get max chat id from db;", "id:", id)
			// INFO: will rewrite first chat
		}
		chat := &models.Chat{
			ID:        id + 1,
			CreatedAt: time.Now(),
			UpdatedAt: time.Now(),
			Agent:     cfg.AssistantRole,
		}
		chat.Name = fmt.Sprintf("%d_%s", chat.ID, cfg.AssistantRole)
		chatMap[chat.Name] = chat
		activeChatName = chat.Name
	}
	chatBody.Messages = history
}

func charToStart(agentName string) bool {
	cc, ok := sysMap[agentName]
	if !ok {
		return false
	}
	applyCharCard(cc)
	return true
}

// func textToMsg(rawMsg string) models.RoleMsg {
// 	msg := models.RoleMsg{}
// 	// system and tool?
// 	if strings.HasPrefix(rawMsg, cfg.AssistantIcon) {
// 		msg.Role = cfg.AssistantRole
// 		msg.Content = strings.TrimPrefix(rawMsg, cfg.AssistantIcon)
// 		return msg
// 	}
// 	if strings.HasPrefix(rawMsg, cfg.UserIcon) {
// 		msg.Role = cfg.UserRole
// 		msg.Content = strings.TrimPrefix(rawMsg, cfg.UserIcon)
// 		return msg
// 	}
// 	return msg
// }

// func textSliceToChat(chat []string) []models.RoleMsg {
// 	resp := make([]models.RoleMsg, len(chat))
// 	for i, rawMsg := range chat {
// 		msg := textToMsg(rawMsg)
// 		resp[i] = msg
// 	}
// 	return resp
// }

func init() {
	cfg = config.LoadConfigOrDefault("config.toml")
	defaultStarter = []models.RoleMsg{
		{Role: "system", Content: basicSysMsg},
		{Role: cfg.AssistantRole, Content: defaultFirstMsg},
	}
	logfile, err := os.OpenFile(cfg.LogFile, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0644)
	if err != nil {
		logger.Error("failed to open log file", "error", err, "filename", cfg.LogFile)
		return
	}
	defaultStarterBytes, err = json.Marshal(defaultStarter)
	if err != nil {
		logger.Error("failed to marshal defaultStarter", "error", err)
		return
	}
	// load cards
	basicCard.Role = cfg.AssistantRole
	toolCard.Role = cfg.AssistantRole
	//
	logger = slog.New(slog.NewTextHandler(logfile, nil))
	store = storage.NewProviderSQL("test.db", logger)
	if store == nil {
		os.Exit(1)
	}
	ragger = rag.New(logger, store, cfg)
	// https://github.com/coreydaley/ggerganov-llama.cpp/blob/master/examples/server/README.md
	// load all chats in memory
	if _, err := loadHistoryChats(); err != nil {
		logger.Error("failed to load chat", "error", err)
		return
	}
	lastChat := loadOldChatOrGetNew()
	logger.Info("loaded history")
	chatBody = &models.ChatBody{
		Model:    "modl_name",
		Stream:   true,
		Messages: lastChat,
	}
	// tempLoad()
}