Fix (rag): epub load

bot.go

@@ -826,8 +826,36 @@ func chatWatcher(ctx context.Context) {
 	}
 }
 
+// inpired by https://github.com/rivo/tview/issues/225
+func showSpinner() {
+	spinners := []string{"⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"}
+	var i int
+	botPersona := cfg.AssistantRole
+	if cfg.WriteNextMsgAsCompletionAgent != "" {
+		botPersona = cfg.WriteNextMsgAsCompletionAgent
+	}
+	for botRespMode || toolRunningMode {
+		time.Sleep(100 * time.Millisecond)
+		spin := i % len(spinners)
+		app.QueueUpdateDraw(func() {
+			if toolRunningMode {
+				textArea.SetTitle(spinners[spin] + " tool")
+			} else if botRespMode {
+				textArea.SetTitle(spinners[spin] + " " + botPersona)
+			} else {
+				textArea.SetTitle(spinners[spin] + " input")
+			}
+		})
+		i++
+	}
+	app.QueueUpdateDraw(func() {
+		textArea.SetTitle("input")
+	})
+}
+
 func chatRound(r *models.ChatRoundReq) error {
 	botRespMode = true
+	go showSpinner()
 	updateStatusLine()
 	botPersona := cfg.AssistantRole
 	if cfg.WriteNextMsgAsCompletionAgent != "" {
@@ -1201,7 +1229,9 @@ func findCall(msg, toolCall string) bool {
 	}
 	// Show tool call progress indicator before execution
 	fmt.Fprintf(textView, "\n[yellow::i][tool: %s...][-:-:-]", fc.Name)
+	toolRunningMode = true
 	resp := callToolWithAgent(fc.Name, fc.Args)
+	toolRunningMode = false
 	toolMsg := string(resp) // Remove the "tool response: " prefix and %+v formatting
 	logger.Info("llm used a tool call", "tool_name", fc.Name, "too_args", fc.Args, "id", fc.ID, "tool_resp", toolMsg)
 	fmt.Fprintf(textView, "%s[-:-:b](%d) <%s>: [-:-:-]\n%s\n",

go.mod

@@ -6,17 +6,19 @@ require (
 	github.com/BurntSushi/toml v1.5.0
 	github.com/GrailFinder/google-translate-tts v0.1.3
 	github.com/GrailFinder/searchagent v0.2.0
+	github.com/PuerkitoBio/goquery v1.11.0
 	github.com/gdamore/tcell/v2 v2.13.2
 	github.com/glebarez/go-sqlite v1.22.0
 	github.com/gopxl/beep/v2 v2.1.1
 	github.com/gordonklaus/portaudio v0.0.0-20250206071425-98a94950218b
 	github.com/jmoiron/sqlx v1.4.0
+	github.com/ledongthuc/pdf v0.0.0-20250511090121-5959a4027728
 	github.com/neurosnap/sentences v1.1.2
 	github.com/rivo/tview v0.42.0
+	github.com/yuin/goldmark v1.4.13
 )
 
 require (
-	github.com/PuerkitoBio/goquery v1.11.0 // indirect
 	github.com/andybalholm/cascadia v1.3.3 // indirect
 	github.com/dustin/go-humanize v1.0.1 // indirect
 	github.com/ebitengine/oto/v3 v3.4.0 // indirect

go.sum

@@ -43,6 +43,8 @@ github.com/hashicorp/golang-lru/v2 v2.0.7 h1:a+bsQ5rvGLjzHuww6tVxozPZFVghXaHOwFs
 github.com/hashicorp/golang-lru/v2 v2.0.7/go.mod h1:QeFd9opnmA6QUJc5vARoKUSoFhyfM2/ZepoAG6RGpeM=
 github.com/jmoiron/sqlx v1.4.0 h1:1PLqN7S1UYp5t4SrVVnt4nUVNemrDAtxlulVe+Qgm3o=
 github.com/jmoiron/sqlx v1.4.0/go.mod h1:ZrZ7UsYB/weZdl2Bxg6jCRO9c3YHl8r3ahlKmRT4JLY=
+github.com/ledongthuc/pdf v0.0.0-20250511090121-5959a4027728 h1:QwWKgMY28TAXaDl+ExRDqGQltzXqN/xypdKP86niVn8=
+github.com/ledongthuc/pdf v0.0.0-20250511090121-5959a4027728/go.mod h1:1fEHWurg7pvf5SG6XNE5Q8UZmOwex51Mkx3SLhrW5B4=
 github.com/lib/pq v1.10.9 h1:YXG7RB+JIjhP29X+OtkiDnYaXQwpS4JEWq7dtCCRUEw=
 github.com/lib/pq v1.10.9/go.mod h1:AlVN5x4E4T544tWzH6hKfbfQvm3HdbOxrmggDNAPY9o=
 github.com/lucasb-eyer/go-colorful v1.3.0 h1:2/yBRLdWBZKrf7gB40FoiKfAWYQ0lqNcbuQwVHXptag=
@@ -67,6 +69,7 @@ github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
 github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
 github.com/stretchr/testify v1.10.0 h1:Xv5erBjTwe/5IxqUQTdXv5kgmIvbHo3QQyRwhJsOfJA=
 github.com/stretchr/testify v1.10.0/go.mod h1:r2ic/lqez/lEtzL7wO/rwa5dbSLXVDPFyf8C91i36aY=
+github.com/yuin/goldmark v1.4.13 h1:fVcFKWvrslecOb/tg+Cc05dkeYx540o0FuFt3nUVDoE=
 github.com/yuin/goldmark v1.4.13/go.mod h1:6yULJ656Px+3vBD8DxQVa3kxgyrAnzto9xy5taEt/CY=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20210921155107-089bfa567519/go.mod h1:GvvjBRRGRdwPK5ydBHafDWAxML/pGHZbMvKqRZ5+Abc=

llm.go

@@ -11,7 +11,6 @@ import (
 
 var imageAttachmentPath string // Global variable to track image attachment for next message
 var lastImg string             // for ctrl+j
-var RAGMsg = "Retrieved context for user's query:\n"
 
 // containsToolSysMsg checks if the toolSysMsg already exists in the chat body
 func containsToolSysMsg() bool {
@@ -142,22 +141,6 @@ func (lcp LCPCompletion) FormMsg(msg, role string, resume bool) (io.Reader, erro
 		newMsg = *processMessageTag(&newMsg)
 		chatBody.Messages = append(chatBody.Messages, newMsg)
 	}
-	// if rag - add as system message to avoid conflicts with tool usage
-	if !resume && cfg.RAGEnabled {
-		um := chatBody.Messages[len(chatBody.Messages)-1].Content
-		logger.Debug("RAG is enabled, preparing RAG context", "user_message", um)
-		ragResp, err := chatRagUse(um)
-		if err != nil {
-			logger.Error("failed to form a rag msg", "error", err)
-			return nil, err
-		}
-		logger.Debug("RAG response received", "response_len", len(ragResp),
-			"response_preview", ragResp[:min(len(ragResp), 100)])
-		// Use system role for RAG context to avoid conflicts with tool usage
-		ragMsg := models.RoleMsg{Role: "system", Content: RAGMsg + ragResp}
-		chatBody.Messages = append(chatBody.Messages, ragMsg)
-		logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages))
-	}
 	// sending description of the tools and how to use them
 	if cfg.ToolUse && !resume && role == cfg.UserRole && !containsToolSysMsg() {
 		chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
@@ -301,23 +284,6 @@ func (op LCPChat) FormMsg(msg, role string, resume bool) (io.Reader, error) {
 		logger.Debug("LCPChat FormMsg: added message to chatBody", "role", newMsg.Role,
 			"content_len", len(newMsg.Content), "message_count_after_add", len(chatBody.Messages))
 	}
-	// if rag - add as system message to avoid conflicts with tool usage
-	if !resume && cfg.RAGEnabled {
-		um := chatBody.Messages[len(chatBody.Messages)-1].Content
-		logger.Debug("LCPChat: RAG is enabled, preparing RAG context", "user_message", um)
-		ragResp, err := chatRagUse(um)
-		if err != nil {
-			logger.Error("LCPChat: failed to form a rag msg", "error", err)
-			return nil, err
-		}
-		logger.Debug("LCPChat: RAG response received",
-			"response_len", len(ragResp), "response_preview", ragResp[:min(len(ragResp), 100)])
-		// Use system role for RAG context to avoid conflicts with tool usage
-		ragMsg := models.RoleMsg{Role: "system", Content: RAGMsg + ragResp}
-		chatBody.Messages = append(chatBody.Messages, ragMsg)
-		logger.Debug("LCPChat: RAG message added to chat body", "role", ragMsg.Role,
-			"rag_content_len", len(ragMsg.Content), "message_count_after_rag", len(chatBody.Messages))
-	}
 	filteredMessages, _ := filterMessagesForCurrentCharacter(chatBody.Messages)
 	// openai /v1/chat does not support custom roles; needs to be user, assistant, system
 	// Add persona suffix to the last user message to indicate who the assistant should reply as
@@ -389,22 +355,6 @@ func (ds DeepSeekerCompletion) FormMsg(msg, role string, resume bool) (io.Reader
 		newMsg = *processMessageTag(&newMsg)
 		chatBody.Messages = append(chatBody.Messages, newMsg)
 	}
-	// if rag - add as system message to avoid conflicts with tool usage
-	if !resume && cfg.RAGEnabled {
-		um := chatBody.Messages[len(chatBody.Messages)-1].Content
-		logger.Debug("DeepSeekerCompletion: RAG is enabled, preparing RAG context", "user_message", um)
-		ragResp, err := chatRagUse(um)
-		if err != nil {
-			logger.Error("DeepSeekerCompletion: failed to form a rag msg", "error", err)
-			return nil, err
-		}
-		logger.Debug("DeepSeekerCompletion: RAG response received",
-			"response_len", len(ragResp), "response_preview", ragResp[:min(len(ragResp), 100)])
-		// Use system role for RAG context to avoid conflicts with tool usage
-		ragMsg := models.RoleMsg{Role: "system", Content: RAGMsg + ragResp}
-		chatBody.Messages = append(chatBody.Messages, ragMsg)
-		logger.Debug("DeepSeekerCompletion: RAG message added to chat body", "message_count", len(chatBody.Messages))
-	}
 	// sending description of the tools and how to use them
 	if cfg.ToolUse && !resume && role == cfg.UserRole && !containsToolSysMsg() {
 		chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
@@ -474,22 +424,6 @@ func (ds DeepSeekerChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
 		newMsg = *processMessageTag(&newMsg)
 		chatBody.Messages = append(chatBody.Messages, newMsg)
 	}
-	// if rag - add as system message to avoid conflicts with tool usage
-	if !resume && cfg.RAGEnabled {
-		um := chatBody.Messages[len(chatBody.Messages)-1].Content
-		logger.Debug("RAG is enabled, preparing RAG context", "user_message", um)
-		ragResp, err := chatRagUse(um)
-		if err != nil {
-			logger.Error("failed to form a rag msg", "error", err)
-			return nil, err
-		}
-		logger.Debug("RAG response received", "response_len", len(ragResp),
-			"response_preview", ragResp[:min(len(ragResp), 100)])
-		// Use system role for RAG context to avoid conflicts with tool usage
-		ragMsg := models.RoleMsg{Role: "system", Content: RAGMsg + ragResp}
-		chatBody.Messages = append(chatBody.Messages, ragMsg)
-		logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages))
-	}
 	// Create copy of chat body with standardized user role
 	filteredMessages, _ := filterMessagesForCurrentCharacter(chatBody.Messages)
 	// Add persona suffix to the last user message to indicate who the assistant should reply as
@@ -552,22 +486,6 @@ func (or OpenRouterCompletion) FormMsg(msg, role string, resume bool) (io.Reader
 		newMsg = *processMessageTag(&newMsg)
 		chatBody.Messages = append(chatBody.Messages, newMsg)
 	}
-	// if rag - add as system message to avoid conflicts with tool usage
-	if !resume && cfg.RAGEnabled {
-		um := chatBody.Messages[len(chatBody.Messages)-1].Content
-		logger.Debug("RAG is enabled, preparing RAG context", "user_message", um)
-		ragResp, err := chatRagUse(um)
-		if err != nil {
-			logger.Error("failed to form a rag msg", "error", err)
-			return nil, err
-		}
-		logger.Debug("RAG response received", "response_len",
-			len(ragResp), "response_preview", ragResp[:min(len(ragResp), 100)])
-		// Use system role for RAG context to avoid conflicts with tool usage
-		ragMsg := models.RoleMsg{Role: "system", Content: RAGMsg + ragResp}
-		chatBody.Messages = append(chatBody.Messages, ragMsg)
-		logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages))
-	}
 	// sending description of the tools and how to use them
 	if cfg.ToolUse && !resume && role == cfg.UserRole && !containsToolSysMsg() {
 		chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
@@ -670,22 +588,6 @@ func (or OpenRouterChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
 		newMsg = *processMessageTag(&newMsg)
 		chatBody.Messages = append(chatBody.Messages, newMsg)
 	}
-	// if rag - add as system message to avoid conflicts with tool usage
-	if !resume && cfg.RAGEnabled {
-		um := chatBody.Messages[len(chatBody.Messages)-1].Content
-		logger.Debug("RAG is enabled, preparing RAG context", "user_message", um)
-		ragResp, err := chatRagUse(um)
-		if err != nil {
-			logger.Error("failed to form a rag msg", "error", err)
-			return nil, err
-		}
-		logger.Debug("RAG response received", "response_len", len(ragResp),
-			"response_preview", ragResp[:min(len(ragResp), 100)])
-		// Use system role for RAG context to avoid conflicts with tool usage
-		ragMsg := models.RoleMsg{Role: "system", Content: RAGMsg + ragResp}
-		chatBody.Messages = append(chatBody.Messages, ragMsg)
-		logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages))
-	}
 	// Create copy of chat body with standardized user role
 	filteredMessages, _ := filterMessagesForCurrentCharacter(chatBody.Messages)
 	// Add persona suffix to the last user message to indicate who the assistant should reply as

main.go

@@ -7,6 +7,7 @@ import (
 var (
 	boolColors        = map[bool]string{true: "green", false: "red"}
 	botRespMode       = false
+	toolRunningMode   = false
 	editMode          = false
 	roleEditMode      = false
 	injectRole        = true

rag/extractors.go

@@ -0,0 +1,184 @@
+package rag
+
+import (
+	"archive/zip"
+	"bytes"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"os/exec"
+	"path"
+	"strings"
+
+	"github.com/PuerkitoBio/goquery"
+	"github.com/ledongthuc/pdf"
+	"github.com/yuin/goldmark"
+	"github.com/yuin/goldmark/extension"
+	"github.com/yuin/goldmark/parser"
+	"github.com/yuin/goldmark/renderer/html"
+)
+
+func ExtractText(fpath string) (string, error) {
+	ext := strings.ToLower(path.Ext(fpath))
+	switch ext {
+	case ".txt":
+		return extractTextFromFile(fpath)
+	case ".md", ".markdown":
+		return extractTextFromMarkdown(fpath)
+	case ".html", ".htm":
+		return extractTextFromHtmlFile(fpath)
+	case ".epub":
+		return extractTextFromEpub(fpath)
+	case ".pdf":
+		return extractTextFromPdf(fpath)
+	default:
+		return "", fmt.Errorf("unsupported file format: %s", ext)
+	}
+}
+
+func extractTextFromFile(fpath string) (string, error) {
+	data, err := os.ReadFile(fpath)
+	if err != nil {
+		return "", err
+	}
+	return string(data), nil
+}
+
+func extractTextFromHtmlFile(fpath string) (string, error) {
+	data, err := os.ReadFile(fpath)
+	if err != nil {
+		return "", err
+	}
+	return extractTextFromHtmlContent(data)
+}
+
+// non utf-8 encoding?
+func extractTextFromHtmlContent(data []byte) (string, error) {
+	doc, err := goquery.NewDocumentFromReader(bytes.NewReader(data))
+	if err != nil {
+		return "", err
+	}
+	// Remove script and style tags
+	doc.Find("script, style, noscript").Each(func(i int, s *goquery.Selection) {
+		s.Remove()
+	})
+	// Get text and clean it
+	text := doc.Text()
+	// Collapse all whitespace (newlines, tabs, multiple spaces) into single spaces
+	cleaned := strings.Join(strings.Fields(text), " ")
+	return cleaned, nil
+}
+
+func extractTextFromMarkdown(fpath string) (string, error) {
+	data, err := os.ReadFile(fpath)
+	if err != nil {
+		return "", err
+	}
+	// Convert markdown to HTML
+	md := goldmark.New(
+		goldmark.WithExtensions(extension.GFM),
+		goldmark.WithParserOptions(parser.WithAutoHeadingID()),
+		goldmark.WithRendererOptions(html.WithUnsafe()), // allow raw HTML if needed
+	)
+	var buf bytes.Buffer
+	if err := md.Convert(data, &buf); err != nil {
+		return "", err
+	}
+	// Now extract text from the resulting HTML (using goquery or similar)
+	return extractTextFromHtmlContent(buf.Bytes())
+}
+
+func extractTextFromEpub(fpath string) (string, error) {
+	r, err := zip.OpenReader(fpath)
+	if err != nil {
+		return "", fmt.Errorf("failed to open epub: %w", err)
+	}
+	defer r.Close()
+
+	var sb strings.Builder
+
+	for _, f := range r.File {
+		ext := strings.ToLower(path.Ext(f.Name))
+		if ext != ".xhtml" && ext != ".html" && ext != ".htm" && ext != ".xml" {
+			continue
+		}
+
+		// Skip manifest, toc, ncx files - they don't contain book content
+		nameLower := strings.ToLower(f.Name)
+		if strings.Contains(nameLower, "toc") || strings.Contains(nameLower, "nav") ||
+			strings.Contains(nameLower, "manifest") || strings.Contains(nameLower, ".opf") ||
+			strings.HasSuffix(nameLower, ".ncx") {
+			continue
+		}
+
+		rc, err := f.Open()
+		if err != nil {
+			continue
+		}
+
+		if sb.Len() > 0 {
+			sb.WriteString("\n\n")
+		}
+		sb.WriteString(f.Name)
+		sb.WriteString("\n")
+
+		buf, readErr := io.ReadAll(rc)
+		rc.Close()
+		if readErr == nil {
+			sb.WriteString(stripHTML(string(buf)))
+		}
+	}
+
+	if sb.Len() == 0 {
+		return "", errors.New("no content extracted from epub")
+	}
+	return sb.String(), nil
+}
+
+func stripHTML(html string) string {
+	var sb strings.Builder
+	inTag := false
+	for _, r := range html {
+		switch r {
+		case '<':
+			inTag = true
+		case '>':
+			inTag = false
+		default:
+			if !inTag {
+				sb.WriteRune(r)
+			}
+		}
+	}
+	return sb.String()
+}
+
+func extractTextFromPdf(fpath string) (string, error) {
+	_, err := exec.LookPath("pdftotext")
+	if err == nil {
+		out, err := exec.Command("pdftotext", "-layout", fpath, "-").Output()
+		if err == nil && len(out) > 0 {
+			return string(out), nil
+		}
+	}
+	return extractTextFromPdfPureGo(fpath)
+}
+
+func extractTextFromPdfPureGo(fpath string) (string, error) {
+	df, r, err := pdf.Open(fpath)
+	if err != nil {
+		return "", fmt.Errorf("failed to open pdf: %w", err)
+	}
+	defer df.Close()
+	textReader, err := r.GetPlainText()
+	if err != nil {
+		return "", fmt.Errorf("failed to extract text from pdf: %w", err)
+	}
+	var buf bytes.Buffer
+	_, err = io.Copy(&buf, textReader)
+	if err != nil {
+		return "", fmt.Errorf("failed to read pdf text: %w", err)
+	}
+	return buf.String(), nil
+}

rag/rag.go

@@ -7,8 +7,9 @@ import (
 	"gf-lt/models"
 	"gf-lt/storage"
 	"log/slog"
-	"os"
 	"path"
+	"regexp"
+	"sort"
 	"strings"
 	"sync"
 
@@ -29,6 +30,7 @@ type RAG struct {
 	cfg      *config.Config
 	embedder Embedder
 	storage  *VectorStorage
+	mu       sync.Mutex
 }
 
 func New(l *slog.Logger, s storage.FullRepo, cfg *config.Config) *RAG {
@@ -53,7 +55,9 @@ func wordCounter(sentence string) int {
 }
 
 func (r *RAG) LoadRAG(fpath string) error {
-	data, err := os.ReadFile(fpath)
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	fileText, err := ExtractText(fpath)
 	if err != nil {
 		return err
 	}
@@ -62,10 +66,7 @@ func (r *RAG) LoadRAG(fpath string) error {
 	case LongJobStatusCh <- LoadedFileRAGStatus:
 	default:
 		r.logger.Warn("LongJobStatusCh channel is full or closed, dropping status message", "message", LoadedFileRAGStatus)
-		// Channel is full or closed, ignore the message to prevent panic
 	}
-
-	fileText := string(data)
 	tokenizer, err := english.NewSentenceTokenizer(nil)
 	if err != nil {
 		return err
@@ -75,19 +76,16 @@ func (r *RAG) LoadRAG(fpath string) error {
 	for i, s := range sentences {
 		sents[i] = s.Text
 	}
-
 	// Group sentences into paragraphs based on word limit
 	paragraphs := []string{}
 	par := strings.Builder{}
 	for i := 0; i < len(sents); i++ {
-		// Only add sentences that aren't empty
 		if strings.TrimSpace(sents[i]) != "" {
 			if par.Len() > 0 {
-				par.WriteString(" ") // Add space between sentences
+				par.WriteString(" ")
 			}
 			par.WriteString(sents[i])
 		}
-
 		if wordCounter(par.String()) > int(r.cfg.RAGWordLimit) {
 			paragraph := strings.TrimSpace(par.String())
 			if paragraph != "" {
@@ -96,7 +94,6 @@ func (r *RAG) LoadRAG(fpath string) error {
 			par.Reset()
 		}
 	}
-
 	// Handle any remaining content in the paragraph buffer
 	if par.Len() > 0 {
 		paragraph := strings.TrimSpace(par.String())
@@ -104,217 +101,82 @@ func (r *RAG) LoadRAG(fpath string) error {
 			paragraphs = append(paragraphs, paragraph)
 		}
 	}
-
 	// Adjust batch size if needed
 	if len(paragraphs) < r.cfg.RAGBatchSize && len(paragraphs) > 0 {
 		r.cfg.RAGBatchSize = len(paragraphs)
 	}
-
 	if len(paragraphs) == 0 {
 		return errors.New("no valid paragraphs found in file")
 	}
-
+	// Process paragraphs in batches synchronously
-	var (
+	batchCount := 0
-		maxChSize = 100
+	for i := 0; i < len(paragraphs); i += r.cfg.RAGBatchSize {
-		left      = 0
+		end := i + r.cfg.RAGBatchSize
-		right     = r.cfg.RAGBatchSize
+		if end > len(paragraphs) {
-		batchCh   = make(chan map[int][]string, maxChSize)
+			end = len(paragraphs)
-		vectorCh  = make(chan []models.VectorRow, maxChSize)
-		errCh     = make(chan error, 1)
-		doneCh    = make(chan struct{})
-		wg        = new(sync.WaitGroup)
-	)
-
-	defer close(doneCh)
-	defer close(errCh)
-	defer close(batchCh)
-
-	// Fill input channel with batches
-	ctn := 0
-	totalParagraphs := len(paragraphs)
-	for {
-		if right > totalParagraphs {
-			batchCh <- map[int][]string{left: paragraphs[left:]}
-			break
 		}
-		batchCh <- map[int][]string{left: paragraphs[left:right]}
+		batch := paragraphs[i:end]
-		left, right = right, right+r.cfg.RAGBatchSize
+		batchCount++
-		ctn++
+		// Filter empty paragraphs
+		nonEmptyBatch := make([]string, 0, len(batch))
+		for _, p := range batch {
+			if strings.TrimSpace(p) != "" {
+				nonEmptyBatch = append(nonEmptyBatch, strings.TrimSpace(p))
 			}
-
-	finishedBatchesMsg := fmt.Sprintf("finished batching batches#: %d; paragraphs: %d; sentences: %d\n", ctn+1, len(paragraphs), len(sents))
-	r.logger.Debug(finishedBatchesMsg)
-	select {
-	case LongJobStatusCh <- finishedBatchesMsg:
-	default:
-		r.logger.Warn("LongJobStatusCh channel is full or closed, dropping status message", "message", finishedBatchesMsg)
-		// Channel is full or closed, ignore the message to prevent panic
 		}
-
+		if len(nonEmptyBatch) == 0 {
-	// Start worker goroutines with WaitGroup
+			continue
-	wg.Add(int(r.cfg.RAGWorkers))
-	for w := 0; w < int(r.cfg.RAGWorkers); w++ {
-		go func(workerID int) {
-			defer wg.Done()
-			r.batchToVectorAsync(workerID, batchCh, vectorCh, errCh, doneCh, path.Base(fpath))
-		}(w)
 		}
-
+		// Embed the batch
-	// Close batchCh to signal workers no more data is coming
+		embeddings, err := r.embedder.EmbedSlice(nonEmptyBatch)
-	close(batchCh)
-
-	// Wait for all workers to finish, then close vectorCh
-	go func() {
-		wg.Wait()
-		close(vectorCh)
-	}()
-
-	// Check for errors from workers - this will block until an error occurs or all workers finish
-	select {
-	case err := <-errCh:
 		if err != nil {
-			r.logger.Error("error during RAG processing", "error", err)
+			r.logger.Error("failed to embed batch", "error", err, "batch", batchCount)
+			select {
+			case LongJobStatusCh <- ErrRAGStatus:
+			default:
+				r.logger.Warn("LongJobStatusCh channel full, dropping message")
+			}
+			return fmt.Errorf("failed to embed batch %d: %w", batchCount, err)
+		}
+		if len(embeddings) != len(nonEmptyBatch) {
+			err := errors.New("embedding count mismatch")
+			r.logger.Error("embedding mismatch", "expected", len(nonEmptyBatch), "got", len(embeddings))
 			return err
 		}
-	default:
+		// Write vectors to storage
-		// No immediate error, continue
+		filename := path.Base(fpath)
+		for j, text := range nonEmptyBatch {
+			vector := models.VectorRow{
+				Embeddings: embeddings[j],
+				RawText:    text,
+				Slug:       fmt.Sprintf("%s_%d_%d", filename, batchCount, j),
+				FileName:   filename,
 			}
-
-	// Write vectors to storage - this will block until vectorCh is closed
-	return r.writeVectors(vectorCh, errCh)
-}
-
-func (r *RAG) writeVectors(vectorCh chan []models.VectorRow, errCh chan error) error {
-	// Use a select to handle both vectorCh and errCh
-	for {
-		select {
-		case err := <-errCh:
-			if err != nil {
-				r.logger.Error("error during RAG processing in writeVectors", "error", err)
-				return err
-			}
-		case batch, ok := <-vectorCh:
-			if !ok {
-				r.logger.Debug("vector channel closed, finished writing vectors")
-				select {
-				case LongJobStatusCh <- FinishedRAGStatus:
-				default:
-					r.logger.Warn("LongJobStatusCh channel is full or closed, dropping status message", "message", FinishedRAGStatus)
-				}
-				return nil
-			}
-			for _, vector := range batch {
 			if err := r.storage.WriteVector(&vector); err != nil {
 				r.logger.Error("failed to write vector to DB", "error", err, "slug", vector.Slug)
 				select {
 				case LongJobStatusCh <- ErrRAGStatus:
 				default:
-						r.logger.Warn("LongJobStatusCh channel is full or closed, dropping status message", "message", ErrRAGStatus)
+					r.logger.Warn("LongJobStatusCh channel full, dropping message")
 				}
-					return err
+				return fmt.Errorf("failed to write vector: %w", err)
 			}
 		}
-			r.logger.Debug("wrote batch to db", "size", len(batch))
+		r.logger.Debug("wrote batch to db", "batch", batchCount, "size", len(nonEmptyBatch))
-		}
+		// Send progress status
-	}
+		statusMsg := fmt.Sprintf("processed batch %d/%d", batchCount, (len(paragraphs)+r.cfg.RAGBatchSize-1)/r.cfg.RAGBatchSize)
-}
-
-func (r *RAG) batchToVectorAsync(id int, inputCh <-chan map[int][]string,
-	vectorCh chan<- []models.VectorRow, errCh chan error, doneCh <-chan struct{}, filename string) {
-	var err error
-
-	defer func() {
-		if err != nil {
-			select {
-			case errCh <- err:
-			default:
-				r.logger.Warn("errCh channel full or closed, skipping error propagation", "worker", id, "error", err)
-			}
-		}
-	}()
-
-	for {
-		select {
-		case <-doneCh:
-			r.logger.Debug("worker received done signal", "worker", id)
-			return
-		case linesMap, ok := <-inputCh:
-			if !ok {
-				r.logger.Debug("input channel closed, worker exiting", "worker", id)
-				return
-			}
-			for leftI, lines := range linesMap {
-				select {
-				case <-doneCh:
-					return
-				default:
-				}
-				if err := r.fetchEmb(lines, errCh, vectorCh, fmt.Sprintf("%s_%d", filename, leftI), filename); err != nil {
-					r.logger.Error("error fetching embeddings", "error", err, "worker", id)
-					return
-				}
-			}
-			r.logger.Debug("processed batch", "worker#", id)
-			statusMsg := fmt.Sprintf("converted to vector; worker#: %d", id)
 		select {
 		case LongJobStatusCh <- statusMsg:
 		default:
-				r.logger.Warn("LongJobStatusCh channel full or closed, dropping status message", "message", statusMsg)
+			r.logger.Warn("LongJobStatusCh channel full, dropping message")
 		}
 	}
+	r.logger.Debug("finished writing vectors", "batches", batchCount)
+	select {
+	case LongJobStatusCh <- FinishedRAGStatus:
+	default:
+		r.logger.Warn("LongJobStatusCh channel is full or closed, dropping status message", "message", FinishedRAGStatus)
 	}
-}
-
-func (r *RAG) fetchEmb(lines []string, errCh chan error, vectorCh chan<- []models.VectorRow, slug, filename string) error {
-	// Filter out empty lines before sending to embedder
-	nonEmptyLines := make([]string, 0, len(lines))
-	for _, line := range lines {
-		trimmed := strings.TrimSpace(line)
-		if trimmed != "" {
-			nonEmptyLines = append(nonEmptyLines, trimmed)
-		}
-	}
-
-	// Skip if no non-empty lines
-	if len(nonEmptyLines) == 0 {
-		// Send empty result but don't error
-		vectorCh <- []models.VectorRow{}
-		return nil
-	}
-
-	embeddings, err := r.embedder.EmbedSlice(nonEmptyLines)
-	if err != nil {
-		r.logger.Error("failed to embed lines", "err", err.Error())
-		errCh <- err
-		return err
-	}
-
-	if len(embeddings) == 0 {
-		err := errors.New("no embeddings returned")
-		r.logger.Error("empty embeddings")
-		errCh <- err
-		return err
-	}
-
-	if len(embeddings) != len(nonEmptyLines) {
-		err := errors.New("mismatch between number of lines and embeddings returned")
-		r.logger.Error("embedding mismatch", "err", err.Error())
-		errCh <- err
-		return err
-	}
-
-	// Create a VectorRow for each line in the batch
-	vectors := make([]models.VectorRow, len(nonEmptyLines))
-	for i, line := range nonEmptyLines {
-		vectors[i] = models.VectorRow{
-			Embeddings: embeddings[i],
-			RawText:    line,
-			Slug:       fmt.Sprintf("%s_%d", slug, i),
-			FileName:   filename,
-		}
-	}
-
-	vectorCh <- vectors
 	return nil
 }
 
@@ -333,3 +195,309 @@ func (r *RAG) ListLoaded() ([]string, error) {
 func (r *RAG) RemoveFile(filename string) error {
 	return r.storage.RemoveEmbByFileName(filename)
 }
+
+var (
+	queryRefinementPattern = regexp.MustCompile(`(?i)(based on my (vector db|vector db|vector database|rags?|past (conversations?|chat|messages?))|from my (files?|documents?|data|information|memory)|search (in|my) (vector db|database|rags?)|rag search for)`)
+	importantKeywords      = []string{"project", "architecture", "code", "file", "chat", "conversation", "topic", "summary", "details", "history", "previous", "my", "user", "me"}
+	stopWords              = []string{"the", "a", "an", "and", "or", "but", "in", "on", "at", "to", "for", "of", "with", "by", "from", "up", "down", "left", "right"}
+)
+
+func (r *RAG) RefineQuery(query string) string {
+	original := query
+	query = strings.TrimSpace(query)
+
+	if len(query) == 0 {
+		return original
+	}
+
+	if len(query) <= 3 {
+		return original
+	}
+
+	query = strings.ToLower(query)
+
+	for _, stopWord := range stopWords {
+		wordPattern := `\b` + stopWord + `\b`
+		re := regexp.MustCompile(wordPattern)
+		query = re.ReplaceAllString(query, "")
+	}
+
+	query = strings.TrimSpace(query)
+
+	if len(query) < 5 {
+		return original
+	}
+
+	if queryRefinementPattern.MatchString(original) {
+		cleaned := queryRefinementPattern.ReplaceAllString(original, "")
+		cleaned = strings.TrimSpace(cleaned)
+		if len(cleaned) >= 5 {
+			return cleaned
+		}
+	}
+
+	query = r.extractImportantPhrases(query)
+
+	if len(query) < 5 {
+		return original
+	}
+
+	return query
+}
+
+func (r *RAG) extractImportantPhrases(query string) string {
+	words := strings.Fields(query)
+
+	var important []string
+	for _, word := range words {
+		word = strings.Trim(word, ".,!?;:'\"()[]{}")
+
+		isImportant := false
+		for _, kw := range importantKeywords {
+			if strings.Contains(strings.ToLower(word), kw) {
+				isImportant = true
+				break
+			}
+		}
+
+		if isImportant || len(word) > 3 {
+			important = append(important, word)
+		}
+	}
+
+	if len(important) == 0 {
+		return query
+	}
+
+	return strings.Join(important, " ")
+}
+
+func (r *RAG) GenerateQueryVariations(query string) []string {
+	variations := []string{query}
+
+	if len(query) < 5 {
+		return variations
+	}
+
+	parts := strings.Fields(query)
+	if len(parts) == 0 {
+		return variations
+	}
+
+	if len(parts) >= 2 {
+		trimmed := strings.Join(parts[:len(parts)-1], " ")
+		if len(trimmed) >= 5 {
+			variations = append(variations, trimmed)
+		}
+	}
+
+	if len(parts) >= 2 {
+		trimmed := strings.Join(parts[1:], " ")
+		if len(trimmed) >= 5 {
+			variations = append(variations, trimmed)
+		}
+	}
+
+	if !strings.HasSuffix(query, " explanation") {
+		variations = append(variations, query+" explanation")
+	}
+	if !strings.HasPrefix(query, "what is ") {
+		variations = append(variations, "what is "+query)
+	}
+	if !strings.HasSuffix(query, " details") {
+		variations = append(variations, query+" details")
+	}
+	if !strings.HasSuffix(query, " summary") {
+		variations = append(variations, query+" summary")
+	}
+
+	return variations
+}
+
+func (r *RAG) RerankResults(results []models.VectorRow, query string) []models.VectorRow {
+	type scoredResult struct {
+		row      models.VectorRow
+		distance float32
+	}
+
+	scored := make([]scoredResult, 0, len(results))
+
+	for i := range results {
+		row := results[i]
+
+		score := float32(0)
+
+		rawTextLower := strings.ToLower(row.RawText)
+		queryLower := strings.ToLower(query)
+
+		if strings.Contains(rawTextLower, queryLower) {
+			score += 10
+		}
+
+		queryWords := strings.Fields(queryLower)
+		matchCount := 0
+		for _, word := range queryWords {
+			if len(word) > 2 && strings.Contains(rawTextLower, word) {
+				matchCount++
+			}
+		}
+		if len(queryWords) > 0 {
+			score += float32(matchCount) / float32(len(queryWords)) * 5
+		}
+
+		if row.FileName == "chat" || strings.Contains(strings.ToLower(row.FileName), "conversation") {
+			score += 3
+		}
+
+		distance := row.Distance - score/100
+
+		scored = append(scored, scoredResult{row: row, distance: distance})
+	}
+
+	sort.Slice(scored, func(i, j int) bool {
+		return scored[i].distance < scored[j].distance
+	})
+
+	unique := make([]models.VectorRow, 0)
+	seen := make(map[string]bool)
+
+	for i := range scored {
+		if !seen[scored[i].row.Slug] {
+			seen[scored[i].row.Slug] = true
+			unique = append(unique, scored[i].row)
+		}
+	}
+
+	if len(unique) > 10 {
+		unique = unique[:10]
+	}
+
+	return unique
+}
+
+func (r *RAG) SynthesizeAnswer(results []models.VectorRow, query string) (string, error) {
+	if len(results) == 0 {
+		return "No relevant information found in the vector database.", nil
+	}
+
+	var contextBuilder strings.Builder
+	contextBuilder.WriteString("User Query: ")
+	contextBuilder.WriteString(query)
+	contextBuilder.WriteString("\n\nRetrieved Context:\n")
+
+	for i, row := range results {
+		contextBuilder.WriteString(fmt.Sprintf("[Source %d: %s]\n", i+1, row.FileName))
+		contextBuilder.WriteString(row.RawText)
+		contextBuilder.WriteString("\n\n")
+	}
+
+	contextBuilder.WriteString("Instructions: ")
+	contextBuilder.WriteString("Based on the retrieved context above, provide a concise, coherent answer to the user's query. ")
+	contextBuilder.WriteString("Extract only the most relevant information. ")
+	contextBuilder.WriteString("If no relevant information is found, state that clearly. ")
+	contextBuilder.WriteString("Cite sources by filename when relevant. ")
+	contextBuilder.WriteString("Do not include unnecessary preamble or explanations.")
+
+	synthesisPrompt := contextBuilder.String()
+
+	emb, err := r.LineToVector(synthesisPrompt)
+	if err != nil {
+		r.logger.Error("failed to embed synthesis prompt", "error", err)
+		return "", err
+	}
+
+	embResp := &models.EmbeddingResp{
+		Embedding: emb,
+		Index:     0,
+	}
+
+	topResults, err := r.SearchEmb(embResp)
+	if err != nil {
+		r.logger.Error("failed to search for synthesis context", "error", err)
+		return "", err
+	}
+
+	if len(topResults) > 0 && topResults[0].RawText != synthesisPrompt {
+		return topResults[0].RawText, nil
+	}
+
+	var finalAnswer strings.Builder
+	finalAnswer.WriteString("Based on the retrieved context:\n\n")
+
+	for i, row := range results {
+		if i >= 5 {
+			break
+		}
+		finalAnswer.WriteString(fmt.Sprintf("- From %s: %s\n", row.FileName, truncateString(row.RawText, 200)))
+	}
+
+	return finalAnswer.String(), nil
+}
+
+func truncateString(s string, maxLen int) string {
+	if len(s) <= maxLen {
+		return s
+	}
+	return s[:maxLen] + "..."
+}
+
+func (r *RAG) Search(query string, limit int) ([]models.VectorRow, error) {
+	refined := r.RefineQuery(query)
+	variations := r.GenerateQueryVariations(refined)
+
+	allResults := make([]models.VectorRow, 0)
+	seen := make(map[string]bool)
+
+	for _, q := range variations {
+		emb, err := r.LineToVector(q)
+		if err != nil {
+			r.logger.Error("failed to embed query variation", "error", err, "query", q)
+			continue
+		}
+
+		embResp := &models.EmbeddingResp{
+			Embedding: emb,
+			Index:     0,
+		}
+
+		results, err := r.SearchEmb(embResp)
+		if err != nil {
+			r.logger.Error("failed to search embeddings", "error", err, "query", q)
+			continue
+		}
+
+		for _, row := range results {
+			if !seen[row.Slug] {
+				seen[row.Slug] = true
+				allResults = append(allResults, row)
+			}
+		}
+	}
+
+	reranked := r.RerankResults(allResults, query)
+
+	if len(reranked) > limit {
+		reranked = reranked[:limit]
+	}
+
+	return reranked, nil
+}
+
+var (
+	ragInstance *RAG
+	ragOnce     sync.Once
+)
+
+func Init(c *config.Config, l *slog.Logger, s storage.FullRepo) error {
+	ragOnce.Do(func() {
+		if c == nil || l == nil || s == nil {
+			return
+		}
+		ragInstance = New(l, s, c)
+	})
+	return nil
+}
+
+func GetInstance() *RAG {
+	return ragInstance
+}

tables.go

@@ -247,9 +247,49 @@ func formatSize(size int64) string {
 	return fmt.Sprintf("%.1f%s", s, units[i])
 }
 
-func makeRAGTable(fileList []string) *tview.Flex {
+type ragFileInfo struct {
-	actions := []string{"load", "delete"}
+	name     string
-	rows, cols := len(fileList), len(actions)+2
+	inRAGDir bool
+	isLoaded bool
+	fullPath string
+}
+
+func makeRAGTable(fileList []string, loadedFiles []string) *tview.Flex {
+	// Build set of loaded files for quick lookup
+	loadedSet := make(map[string]bool)
+	for _, f := range loadedFiles {
+		loadedSet[f] = true
+	}
+
+	// Build merged list: files from ragdir + orphaned files from DB
+	ragFiles := make([]ragFileInfo, 0, len(fileList)+len(loadedFiles))
+	seen := make(map[string]bool)
+
+	// Add files from ragdir
+	for _, f := range fileList {
+		ragFiles = append(ragFiles, ragFileInfo{
+			name:     f,
+			inRAGDir: true,
+			isLoaded: loadedSet[f],
+			fullPath: path.Join(cfg.RAGDir, f),
+		})
+		seen[f] = true
+	}
+
+	// Add orphaned files (in DB but not in ragdir)
+	for _, f := range loadedFiles {
+		if !seen[f] {
+			ragFiles = append(ragFiles, ragFileInfo{
+				name:     f,
+				inRAGDir: false,
+				isLoaded: true,
+				fullPath: "",
+			})
+		}
+	}
+
+	rows := len(ragFiles)
+	cols := 4 // File Name | Preview | Action | Delete
 	fileTable := tview.NewTable().
 		SetBorders(true)
 	longStatusView := tview.NewTextView()
@@ -273,7 +313,7 @@ func makeRAGTable(fileList []string) *tview.Flex {
 			SetAlign(tview.AlignCenter).
 			SetSelectable(false))
 	fileTable.SetCell(0, 2,
-		tview.NewTableCell("Load").
+		tview.NewTableCell("Load/Unload").
 			SetTextColor(tcell.ColorWhite).
 			SetAlign(tview.AlignCenter).
 			SetSelectable(false))
@@ -284,18 +324,29 @@ func makeRAGTable(fileList []string) *tview.Flex {
 			SetSelectable(false))
 	// Add the file rows starting from row 1
 	for r := 0; r < rows; r++ {
+		f := ragFiles[r]
 		for c := 0; c < cols; c++ {
 			color := tcell.ColorWhite
 			switch {
 			case c == 0:
+				displayName := f.name
+				if !f.inRAGDir {
+					displayName = f.name + " (orphaned)"
+				}
 				fileTable.SetCell(r+1, c,
-					tview.NewTableCell(fileList[r]).
+					tview.NewTableCell(displayName).
 						SetTextColor(color).
 						SetAlign(tview.AlignCenter).
 						SetSelectable(false))
 			case c == 1:
-				fpath := path.Join(cfg.RAGDir, fileList[r])
+				if !f.inRAGDir {
-				if fi, err := os.Stat(fpath); err == nil {
+					// Orphaned file - no preview available
+					fileTable.SetCell(r+1, c,
+						tview.NewTableCell("not in ragdir").
+							SetTextColor(tcell.ColorYellow).
+							SetAlign(tview.AlignCenter).
+							SetSelectable(false))
+				} else if fi, err := os.Stat(f.fullPath); err == nil {
 					size := fi.Size()
 					modTime := fi.ModTime()
 					preview := fmt.Sprintf("%s | %s", formatSize(size), modTime.Format("2006-01-02 15:04"))
@@ -312,11 +363,27 @@ func makeRAGTable(fileList []string) *tview.Flex {
 							SetSelectable(false))
 				}
 			case c == 2:
+				actionText := "load"
+				if f.isLoaded {
+					actionText = "unload"
+				}
+				if !f.inRAGDir {
+					// Orphaned file - can only unload
+					actionText = "unload"
+				}
 				fileTable.SetCell(r+1, c,
-					tview.NewTableCell("load").
+					tview.NewTableCell(actionText).
 						SetTextColor(color).
 						SetAlign(tview.AlignCenter))
-			default:
+			case c == 3:
+				if !f.inRAGDir {
+					// Orphaned file - cannot delete from ragdir (not there)
+					fileTable.SetCell(r+1, c,
+						tview.NewTableCell("-").
+							SetTextColor(tcell.ColorDarkGray).
+							SetAlign(tview.AlignCenter).
+							SetSelectable(false))
+				} else {
 					fileTable.SetCell(r+1, c,
 						tview.NewTableCell("delete").
 							SetTextColor(color).
@@ -324,6 +391,7 @@ func makeRAGTable(fileList []string) *tview.Flex {
 				}
 			}
 		}
+	}
 	errCh := make(chan error, 1) // why?
 	go func() {
 		defer pages.RemovePage(RAGPage)
@@ -376,25 +444,51 @@ func makeRAGTable(fileList []string) *tview.Flex {
 			pages.RemovePage(RAGPage)
 			return
 		}
-		// For file rows, get the filename (row index - 1 because of the exit row at index 0)
+		// For file rows, get the file info (row index - 1 because of the exit row at index 0)
-		fpath := fileList[row-1] // -1 to account for the exit row at index 0
+		f := ragFiles[row-1]
-		// notification := fmt.Sprintf("chat: %s; action: %s", fpath, tc.Text)
+		// Handle "-" case (orphaned file with no delete option)
+		if tc.Text == "-" {
+			pages.RemovePage(RAGPage)
+			return
+		}
 		switch tc.Text {
 		case "load":
-			fpath = path.Join(cfg.RAGDir, fpath)
+			fpath := path.Join(cfg.RAGDir, f.name)
 			longStatusView.SetText("clicked load")
 			go func() {
 				if err := ragger.LoadRAG(fpath); err != nil {
 					logger.Error("failed to embed file", "chat", fpath, "error", err)
 					_ = notifyUser("RAG", "failed to embed file; error: "+err.Error())
-					errCh <- err
+					app.QueueUpdate(func() {
-					// pages.RemovePage(RAGPage)
+						pages.RemovePage(RAGPage)
+					})
 					return
 				}
+				_ = notifyUser("RAG", "file loaded successfully")
+				app.QueueUpdate(func() {
+					pages.RemovePage(RAGPage)
+				})
+			}()
+			return
+		case "unload":
+			longStatusView.SetText("clicked unload")
+			go func() {
+				if err := ragger.RemoveFile(f.name); err != nil {
+					logger.Error("failed to unload file from RAG", "filename", f.name, "error", err)
+					_ = notifyUser("RAG", "failed to unload file; error: "+err.Error())
+					app.QueueUpdate(func() {
+						pages.RemovePage(RAGPage)
+					})
+					return
+				}
+				_ = notifyUser("RAG", "file unloaded successfully")
+				app.QueueUpdate(func() {
+					pages.RemovePage(RAGPage)
+				})
 			}()
 			return
 		case "delete":
-			fpath = path.Join(cfg.RAGDir, fpath)
+			fpath := path.Join(cfg.RAGDir, f.name)
 			if err := os.Remove(fpath); err != nil {
 				logger.Error("failed to delete file", "filename", fpath, "error", err)
 				return

tools.go

@@ -16,6 +16,7 @@ import (
 	"sync"
 	"time"
 
+	"gf-lt/rag"
 	"github.com/GrailFinder/searchagent/searcher"
 )
 
@@ -58,9 +59,9 @@ Your current tools:
 "when_to_use": "when asked to search the web for information; returns clean summary without html,css and other web elements; limit is optional (default 3)"
 },
 {
-"name":"websearch_raw",
+"name":"rag_search",
 "args": ["query", "limit"],
-"when_to_use": "when asked to search the web for information; returns raw data as is without processing; limit is optional (default 3)"
+"when_to_use": "when asked to search the local document database for information; performs query refinement, semantic search, reranking, and synthesis; returns clean summary with sources; limit is optional (default 3)"
 },
 {
 "name":"read_url",
@@ -146,6 +147,7 @@ under the topic: Adam's number is stored:
 After that you are free to respond to the user.
 `
 	webSearchSysPrompt = `Summarize the web search results, extracting key information and presenting a concise answer. Provide sources and URLs where relevant.`
+	ragSearchSysPrompt = `Synthesize the document search results, extracting key information and presenting a concise answer. Provide sources and document IDs where relevant.`
 	readURLSysPrompt   = `Extract and summarize the content from the webpage. Provide key information, main points, and any relevant details.`
 	summarySysPrompt   = `Please provide a concise summary of the following conversation. Focus on key points, decisions, and actions. Provide only the summary, no additional commentary.`
 	basicCard          = &models.CharCard{
@@ -170,6 +172,10 @@ func init() {
 		panic("failed to init seachagent; error: " + err.Error())
 	}
 	WebSearcher = sa
+
+	if err := rag.Init(cfg, logger, store); err != nil {
+		logger.Warn("failed to init rag; rag_search tool will not be available", "error", err)
+	}
 }
 
 // getWebAgentClient returns a singleton AgentClient for web agents.
@@ -196,6 +202,8 @@ func getWebAgentClient() *agent.AgentClient {
 func registerWebAgents() {
 	webAgentsOnce.Do(func() {
 		client := getWebAgentClient()
+		// Register rag_search agent
+		agent.Register("rag_search", agent.NewWebAgentB(client, ragSearchSysPrompt))
 		// Register websearch agent
 		agent.Register("websearch", agent.NewWebAgentB(client, webSearchSysPrompt))
 		// Register read_url agent
@@ -239,6 +247,48 @@ func websearch(args map[string]string) []byte {
 	return data
 }
 
+// rag search (searches local document database)
+func ragsearch(args map[string]string) []byte {
+	query, ok := args["query"]
+	if !ok || query == "" {
+		msg := "query not provided to rag_search tool"
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	limitS, ok := args["limit"]
+	if !ok || limitS == "" {
+		limitS = "3"
+	}
+	limit, err := strconv.Atoi(limitS)
+	if err != nil || limit == 0 {
+		logger.Warn("ragsearch limit; passed bad value; setting to default (3)",
+			"limit_arg", limitS, "error", err)
+		limit = 3
+	}
+
+	ragInstance := rag.GetInstance()
+	if ragInstance == nil {
+		msg := "rag not initialized; rag_search tool is not available"
+		logger.Error(msg)
+		return []byte(msg)
+	}
+
+	results, err := ragInstance.Search(query, limit)
+	if err != nil {
+		msg := "rag search failed; error: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+
+	data, err := json.Marshal(results)
+	if err != nil {
+		msg := "failed to marshal rag search result; error: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	return data
+}
+
 // web search raw (returns raw data without processing)
 func websearchRaw(args map[string]string) []byte {
 	// make http request return bytes
@@ -997,6 +1047,7 @@ var fnMap = map[string]fnSig{
 	"recall":            recall,
 	"recall_topics":     recallTopics,
 	"memorise":          memorise,
+	"rag_search":        ragsearch,
 	"websearch":         websearch,
 	"websearch_raw":     websearchRaw,
 	"read_url":          readURL,
@@ -1033,6 +1084,28 @@ func callToolWithAgent(name string, args map[string]string) []byte {
 
 // openai style def
 var baseTools = []models.Tool{
+	// rag_search
+	models.Tool{
+		Type: "function",
+		Function: models.ToolFunc{
+			Name:        "rag_search",
+			Description: "Search local document database given query, limit of sources (default 3). Performs query refinement, semantic search, reranking, and synthesis.",
+			Parameters: models.ToolFuncParams{
+				Type:     "object",
+				Required: []string{"query", "limit"},
+				Properties: map[string]models.ToolArgProps{
+					"query": models.ToolArgProps{
+						Type:        "string",
+						Description: "search query",
+					},
+					"limit": models.ToolArgProps{
+						Type:        "string",
+						Description: "limit of the document results",
+					},
+				},
+			},
+		},
+	},
 	// websearch
 	models.Tool{
 		Type: "function",

tui.go

@@ -913,6 +913,7 @@ func init() {
 					return nil
 				}
 			}
+			// Get files from ragdir
 			fileList := []string{}
 			for _, f := range files {
 				if f.IsDir() {
@@ -920,22 +921,14 @@ func init() {
 				}
 				fileList = append(fileList, f.Name())
 			}
-			chatRAGTable := makeRAGTable(fileList)
+			// Get loaded files from vector DB
-			pages.AddPage(RAGPage, chatRAGTable, true, true)
+			loadedFiles, err := ragger.ListLoaded()
-			return nil
-		}
-		if event.Key() == tcell.KeyCtrlY { // Use Ctrl+Y to list loaded RAG files
-			// List files already loaded into the RAG system
-			fileList, err := ragger.ListLoaded()
 			if err != nil {
 				logger.Error("failed to list loaded RAG files", "error", err)
-				if notifyerr := notifyUser("failed to list RAG files", err.Error()); notifyerr != nil {
+				loadedFiles = []string{} // Continue with empty list on error
-					logger.Error("failed to send notification", "error", notifyerr)
 			}
-				return nil
+			chatRAGTable := makeRAGTable(fileList, loadedFiles)
-			}
+			pages.AddPage(RAGPage, chatRAGTable, true, true)
-			chatLoadedRAGTable := makeLoadedRAGTable(fileList)
-			pages.AddPage(RAGLoadedPage, chatLoadedRAGTable, true, true)
 			return nil
 		}
 		if event.Key() == tcell.KeyRune && event.Modifiers() == tcell.ModAlt && event.Rune() == '1' {