Fix: linter complaints

logic requires reference to config

Makefile

@@ -1,4 +1,4 @@
-.PHONY: setconfig run lint install-linters setup-whisper build-whisper download-whisper-model docker-up docker-down docker-logs noextra-run installdelve checkdelve
+.PHONY: setconfig run lint lintall install-linters setup-whisper build-whisper download-whisper-model docker-up docker-down docker-logs noextra-run installdelve checkdelve
 
 run: setconfig
 	go build -tags extra -o gf-lt && ./gf-lt
@@ -25,7 +25,10 @@ install-linters: ## Install additional linters (noblanks)
 	go install github.com/GrailFinder/noblanks-linter/cmd/noblanks@latest
 
 lint: ## Run linters. Use make install-linters first.
-	golangci-lint run -c .golangci.yml ./...; noblanks ./...
+	golangci-lint run -c .golangci.yml ./...
+
+lintall: lint
+	noblanks ./...
 
 # Whisper STT Setup (in batteries directory)
 setup-whisper: build-whisper download-whisper-model

agent/request.go

@@ -71,8 +71,8 @@ func (ag *AgentClient) buildRequest(sysprompt, msg string) ([]byte, error) {
 	// Build prompt for completion endpoints
 	if isCompletion {
 		var sb strings.Builder
-		for _, m := range messages {
+		for i := range messages {
-			sb.WriteString(m.ToPrompt())
+			sb.WriteString(messages[i].ToPrompt())
 			sb.WriteString("\n")
 		}
 		prompt := strings.TrimSpace(sb.String())

bot.go

@@ -66,6 +66,8 @@ var (
 	LocalModels = []string{}
 )
 
+var thinkBlockRE = regexp.MustCompile(`(?s)<think>.*?</think>`)
+
 // parseKnownToTag extracts known_to list from content using configured tag.
 // Returns cleaned content and list of character names.
 func parseKnownToTag(content string) []string {
@@ -136,6 +138,9 @@ func processMessageTag(msg *models.RoleMsg) *models.RoleMsg {
 // filterMessagesForCharacter returns messages visible to the specified character.
 // If CharSpecificContextEnabled is false, returns all messages.
 func filterMessagesForCharacter(messages []models.RoleMsg, character string) []models.RoleMsg {
+	if strings.Contains(cfg.CurrentAPI, "chat") {
+		return messages
+	}
 	if cfg == nil || !cfg.CharSpecificContextEnabled || character == "" {
 		return messages
 	}
@@ -143,97 +148,67 @@ func filterMessagesForCharacter(messages []models.RoleMsg, character string) []m
 		return messages
 	}
 	filtered := make([]models.RoleMsg, 0, len(messages))
-	for _, msg := range messages {
+	for i := range messages {
 		// If KnownTo is nil or empty, message is visible to all
 		// system msg cannot be filtered
-		if len(msg.KnownTo) == 0 || msg.Role == "system" {
+		if len(messages[i].KnownTo) == 0 || messages[i].Role == "system" {
-			filtered = append(filtered, msg)
+			filtered = append(filtered, messages[i])
 			continue
 		}
-		if slices.Contains(msg.KnownTo, character) {
+		if slices.Contains(messages[i].KnownTo, character) {
 			// Check if character is in KnownTo lis
-			filtered = append(filtered, msg)
+			filtered = append(filtered, messages[i])
 		}
 	}
 	return filtered
 }
 
-func cleanToolCalls(messages []models.RoleMsg) []models.RoleMsg {
-	// If AutoCleanToolCallsFromCtx is false, keep tool call messages in context
-	if cfg != nil && !cfg.AutoCleanToolCallsFromCtx {
-		return consolidateAssistantMessages(messages)
-	}
-	cleaned := make([]models.RoleMsg, 0, len(messages))
-	for i, msg := range messages {
-		// recognize the message as the tool call and remove it
-		// tool call in last msg should stay
-		if msg.ToolCallID == "" || i == len(messages)-1 {
-			cleaned = append(cleaned, msg)
-		}
-	}
-	return consolidateAssistantMessages(cleaned)
-}
-
-// consolidateAssistantMessages merges consecutive assistant messages into a single message
 func consolidateAssistantMessages(messages []models.RoleMsg) []models.RoleMsg {
 	if len(messages) == 0 {
 		return messages
 	}
-	consolidated := make([]models.RoleMsg, 0, len(messages))
+	result := make([]models.RoleMsg, 0, len(messages))
-	currentAssistantMsg := models.RoleMsg{}
+	for i := range messages {
-	isBuildingAssistantMsg := false
+		// Non-assistant messages are appended as-is
-	for i := 0; i < len(messages); i++ {
+		if messages[i].Role != cfg.AssistantRole {
-		msg := messages[i]
+			result = append(result, messages[i])
-		// assistant role only
+			continue
-		if msg.Role == cfg.AssistantRole {
+		}
-			// If this is an assistant message, start or continue building
+		// Assistant message: start a new block or merge with the last one
-			if !isBuildingAssistantMsg {
+		if len(result) == 0 || result[len(result)-1].Role != cfg.AssistantRole {
-				// Start accumulating assistant message
+			// First assistant in a block: append a copy (avoid mutating input)
-				currentAssistantMsg = msg.Copy()
+			result = append(result, messages[i].Copy())
-				isBuildingAssistantMsg = true
+			continue
-			} else {
+		}
-				// Continue accumulating - append content to the current assistant message
+		// Merge with the last assistant message
-				if currentAssistantMsg.IsContentParts() || msg.IsContentParts() {
+		last := &result[len(result)-1]
-					// Handle structured content
+		// If either message has structured content, unify to ContentParts
-					if !currentAssistantMsg.IsContentParts() {
+		if last.IsContentParts() || messages[i].IsContentParts() {
-						// Preserve the original ToolCallID before conversion
+			// Convert last to ContentParts if needed, preserving ToolCallID
-						originalToolCallID := currentAssistantMsg.ToolCallID
+			if !last.IsContentParts() {
-						// Convert existing content to content parts
+				toolCallID := last.ToolCallID
-						currentAssistantMsg = models.NewMultimodalMsg(currentAssistantMsg.Role, []interface{}{models.TextContentPart{Type: "text", Text: currentAssistantMsg.Content}})
+				*last = models.NewMultimodalMsg(last.Role, []interface{}{
-						// Restore the original ToolCallID to preserve tool call linking
+					models.TextContentPart{Type: "text", Text: last.Content},
-						currentAssistantMsg.ToolCallID = originalToolCallID
+				})
-					}
+				last.ToolCallID = toolCallID
-					if msg.IsContentParts() {
+			}
-						currentAssistantMsg.ContentParts = append(currentAssistantMsg.ContentParts, msg.GetContentParts()...)
+			// Add current message's content to last
-					} else if msg.Content != "" {
+			if messages[i].IsContentParts() {
-						currentAssistantMsg.AddTextPart(msg.Content)
+				last.ContentParts = append(last.ContentParts, messages[i].GetContentParts()...)
-					}
+			} else if messages[i].Content != "" {
-				} else {
+				last.AddTextPart(messages[i].Content)
-					// Simple string content
-					if currentAssistantMsg.Content != "" {
-						currentAssistantMsg.Content += "\n" + msg.Content
-					} else {
-						currentAssistantMsg.Content = msg.Content
-					}
-					// ToolCallID is already preserved since we're not creating a new message object when just concatenating content
-				}
 			}
 		} else {
-			// This is not an assistant message
+			// Both simple strings: concatenate with newline
-			// If we were building an assistant message, add it to the result
+			if last.Content != "" && messages[i].Content != "" {
-			if isBuildingAssistantMsg {
+				last.Content += "\n" + messages[i].Content
-				consolidated = append(consolidated, currentAssistantMsg)
+			} else if messages[i].Content != "" {
-				isBuildingAssistantMsg = false
+				last.Content = messages[i].Content
 			}
-			// Add the non-assistant message
+			// ToolCallID is already preserved in last
-			consolidated = append(consolidated, msg)
 		}
 	}
-	// Don't forget the last assistant message if we were building one
+	return result
-	if isBuildingAssistantMsg {
-		consolidated = append(consolidated, currentAssistantMsg)
-	}
-	return consolidated
 }
 
 // GetLogLevel returns the current log level as a string
@@ -404,22 +379,22 @@ func fetchLCPModels() ([]string, error) {
 
 // fetchLCPModelsWithLoadStatus returns models with "(loaded)" indicator for loaded models
 func fetchLCPModelsWithLoadStatus() ([]string, error) {
-	models, err := fetchLCPModelsWithStatus()
+	modelList, err := fetchLCPModelsWithStatus()
 	if err != nil {
 		return nil, err
 	}
-	result := make([]string, 0, len(models.Data))
+	result := make([]string, 0, len(modelList.Data))
 	li := 0 // loaded index
-	for i, m := range models.Data {
+	for i, m := range modelList.Data {
 		modelName := m.ID
 		if m.Status.Value == "loaded" {
-			modelName = "(loaded) " + modelName
+			modelName = models.LoadedMark + modelName
 			li = i
 		}
 		result = append(result, modelName)
 	}
 	if li == 0 {
-		return result, nil // no loaded models
+		return result, nil // no loaded modelList
 	}
 	loadedModel := result[li]
 	result = append(result[:li], result[li+1:]...)
@@ -458,6 +433,33 @@ func isModelLoaded(modelID string) (bool, error) {
 	return false, nil
 }
 
+func ModelHasVision(api, modelID string) bool {
+	switch {
+	case strings.Contains(api, "deepseek"):
+		return false
+	case strings.Contains(api, "openrouter"):
+		resp, err := http.Get("https://openrouter.ai/api/v1/models")
+		if err != nil {
+			logger.Warn("failed to fetch OR models for vision check", "error", err)
+			return false
+		}
+		defer resp.Body.Close()
+		orm := &models.ORModels{}
+		if err := json.NewDecoder(resp.Body).Decode(orm); err != nil {
+			logger.Warn("failed to decode OR models for vision check", "error", err)
+			return false
+		}
+		return orm.HasVision(modelID)
+	default:
+		models, err := fetchLCPModelsWithStatus()
+		if err != nil {
+			logger.Warn("failed to fetch LCP models for vision check", "error", err)
+			return false
+		}
+		return models.HasVision(modelID)
+	}
+}
+
 // monitorModelLoad starts a goroutine that periodically checks if the specified model is loaded.
 func monitorModelLoad(modelID string) {
 	go func() {
@@ -777,7 +779,7 @@ func showSpinner() {
 		botPersona = cfg.WriteNextMsgAsCompletionAgent
 	}
 	for botRespMode || toolRunningMode {
-		time.Sleep(100 * time.Millisecond)
+		time.Sleep(400 * time.Millisecond)
 		spin := i % len(spinners)
 		app.QueueUpdateDraw(func() {
 			switch {
@@ -960,7 +962,9 @@ out:
 	if err := updateStorageChat(activeChatName, chatBody.Messages); err != nil {
 		logger.Warn("failed to update storage", "error", err, "name", activeChatName)
 	}
-	if findCall(respText.String(), toolResp.String()) {
+	// Strip think blocks before parsing for tool calls
+	respTextNoThink := thinkBlockRE.ReplaceAllString(respText.String(), "")
+	if findCall(respTextNoThink, toolResp.String()) {
 		return nil
 	}
 	// Check if this message was sent privately to specific characters
@@ -982,7 +986,7 @@ func cleanChatBody() {
 	}
 	// Tool request cleaning is now configurable via AutoCleanToolCallsFromCtx (default false)
 	// /completion msg where part meant for user and other part tool call
-	chatBody.Messages = cleanToolCalls(chatBody.Messages)
+	// chatBody.Messages = cleanToolCalls(chatBody.Messages)
 	chatBody.Messages = consolidateAssistantMessages(chatBody.Messages)
 }
 
@@ -1096,22 +1100,38 @@ func findCall(msg, toolCall string) bool {
 		}
 		lastToolCall.Args = openAIToolMap
 		fc = lastToolCall
-		// Set lastToolCall.ID from parsed tool call ID if available
+		// NOTE: We do NOT override lastToolCall.ID from arguments.
-		if len(openAIToolMap) > 0 {
+		// The ID should come from the streaming response (chunk.ToolID) set earlier.
-			if id, exists := openAIToolMap["id"]; exists {
+		// Some tools like todo_create have "id" in their arguments which is NOT the tool call ID.
-				lastToolCall.ID = id
-			}
-		}
 	} else {
 		jsStr := toolCallRE.FindString(msg)
 		if jsStr == "" { // no tool call case
 			return false
 		}
-		prefix := "__tool_call__\n"
+		// Remove prefix/suffix with flexible whitespace handling
-		suffix := "\n__tool_call__"
+		jsStr = strings.TrimSpace(jsStr)
-		jsStr = strings.TrimSuffix(strings.TrimPrefix(jsStr, prefix), suffix)
+		jsStr = strings.TrimPrefix(jsStr, "__tool_call__")
+		jsStr = strings.TrimSuffix(jsStr, "__tool_call__")
+		jsStr = strings.TrimSpace(jsStr)
 		// HTML-decode the JSON string to handle encoded characters like &lt; -> <=
 		decodedJsStr := html.UnescapeString(jsStr)
+		// Try to find valid JSON bounds (first { to last })
+		start := strings.Index(decodedJsStr, "{")
+		end := strings.LastIndex(decodedJsStr, "}")
+		if start == -1 || end == -1 || end <= start {
+			logger.Error("failed to find valid JSON in tool call", "json_string", decodedJsStr)
+			toolResponseMsg := models.RoleMsg{
+				Role:    cfg.ToolRole,
+				Content: "Error processing tool call: no valid JSON found. Please check the JSON format.",
+			}
+			chatBody.Messages = append(chatBody.Messages, toolResponseMsg)
+			crr := &models.ChatRoundReq{
+				Role: cfg.AssistantRole,
+			}
+			chatRoundChan <- crr
+			return true
+		}
+		decodedJsStr = decodedJsStr[start : end+1]
 		var err error
 		fc, err = unmarshalFuncCall(decodedJsStr)
 		if err != nil {
@@ -1138,14 +1158,18 @@ func findCall(msg, toolCall string) bool {
 		lastToolCall.Args = fc.Args
 	}
 	// we got here => last msg recognized as a tool call (correct or not)
-	// make sure it has ToolCallID
+	// Use the tool call ID from streaming response (lastToolCall.ID)
-	if chatBody.Messages[len(chatBody.Messages)-1].ToolCallID == "" {
+	// Don't generate random ID - the ID should match between assistant message and tool response
-		// Tool call IDs should be alphanumeric strings with length 9!
+	lastMsgIdx := len(chatBody.Messages) - 1
-		chatBody.Messages[len(chatBody.Messages)-1].ToolCallID = randString(9)
+	if lastToolCall.ID != "" {
+		chatBody.Messages[lastMsgIdx].ToolCallID = lastToolCall.ID
 	}
-	// Ensure lastToolCall.ID is set, fallback to assistant message's ToolCallID
+	// Store tool call info in the assistant message
-	if lastToolCall.ID == "" {
+	// Convert Args map to JSON string for storage
-		lastToolCall.ID = chatBody.Messages[len(chatBody.Messages)-1].ToolCallID
+	chatBody.Messages[lastMsgIdx].ToolCall = &models.ToolCall{
+		ID:   lastToolCall.ID,
+		Name: lastToolCall.Name,
+		Args: mapToString(lastToolCall.Args),
 	}
 	// call a func
 	_, ok := fnMap[fc.Name]
@@ -1175,16 +1199,61 @@ func findCall(msg, toolCall string) bool {
 	toolRunningMode = true
 	resp := callToolWithAgent(fc.Name, fc.Args)
 	toolRunningMode = false
-	toolMsg := string(resp) // Remove the "tool response: " prefix and %+v formatting
+	toolMsg := string(resp)
 	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",
-		"\n\n", len(chatBody.Messages), cfg.ToolRole, toolMsg)
 	// Create tool response message with the proper tool_call_id
-	toolResponseMsg := models.RoleMsg{
+	// Mark shell commands as always visible
-		Role:       cfg.ToolRole,
+	isShellCommand := fc.Name == "execute_command"
-		Content:    toolMsg,
+	// Check if response is multimodal content (image)
-		ToolCallID: lastToolCall.ID, // Use the stored tool call ID
+	var toolResponseMsg models.RoleMsg
+	if strings.HasPrefix(strings.TrimSpace(toolMsg), `{"type":"multimodal_content"`) {
+		// Parse multimodal content response
+		multimodalResp := models.MultimodalToolResp{}
+		if err := json.Unmarshal([]byte(toolMsg), &multimodalResp); err == nil && multimodalResp.Type == "multimodal_content" {
+			// Create RoleMsg with ContentParts
+			var contentParts []any
+			for _, part := range multimodalResp.Parts {
+				partType := part["type"]
+				switch partType {
+				case "text":
+					contentParts = append(contentParts, models.TextContentPart{Type: "text", Text: part["text"]})
+				case "image_url":
+					contentParts = append(contentParts, models.ImageContentPart{
+						Type: "image_url",
+						ImageURL: struct {
+							URL string `json:"url"`
+						}{URL: part["url"]},
+					})
+				default:
+					continue
+				}
+			}
+			toolResponseMsg = models.RoleMsg{
+				Role:            cfg.ToolRole,
+				ContentParts:    contentParts,
+				HasContentParts: true,
+				ToolCallID:      lastToolCall.ID,
+				IsShellCommand:  isShellCommand,
+			}
+		} else {
+			// Fallback to regular content
+			toolResponseMsg = models.RoleMsg{
+				Role:           cfg.ToolRole,
+				Content:        toolMsg,
+				ToolCallID:     lastToolCall.ID,
+				IsShellCommand: isShellCommand,
+			}
+		}
+	} else {
+		toolResponseMsg = models.RoleMsg{
+			Role:           cfg.ToolRole,
+			Content:        toolMsg,
+			ToolCallID:     lastToolCall.ID,
+			IsShellCommand: isShellCommand,
+		}
 	}
+	fmt.Fprintf(textView, "%s[-:-:b](%d) <%s>: [-:-:-]\n%s\n",
+		"\n\n", len(chatBody.Messages), cfg.ToolRole, toolResponseMsg.GetText())
 	chatBody.Messages = append(chatBody.Messages, toolResponseMsg)
 	logger.Debug("findCall: added actual tool response", "role", toolResponseMsg.Role, "content_len", len(toolResponseMsg.Content), "tool_call_id", toolResponseMsg.ToolCallID, "message_count_after_add", len(chatBody.Messages))
 	// Clear the stored tool call ID after using it
@@ -1200,12 +1269,42 @@ func findCall(msg, toolCall string) bool {
 
 func chatToTextSlice(messages []models.RoleMsg, showSys bool) []string {
 	resp := make([]string, len(messages))
-	for i, msg := range messages {
+	for i := range messages {
-		// INFO: skips system msg and tool msg
+		icon := fmt.Sprintf("[-:-:b](%d) <%s>:[-:-:-]", i, messages[i].Role)
-		if !showSys && (msg.Role == cfg.ToolRole || msg.Role == "system") {
+		// Handle tool call indicators (assistant messages with tool call but empty content)
+		if messages[i].Role == cfg.AssistantRole && messages[i].ToolCall != nil && messages[i].ToolCall.ID != "" {
+			// This is a tool call indicator - show collapsed
+			if toolCollapsed {
+				toolName := messages[i].ToolCall.Name
+				resp[i] = strings.ReplaceAll(fmt.Sprintf("%s\n%s\n[yellow::i][tool call: %s (press Ctrl+T to expand)][-:-:-]\n", icon, messages[i].GetText(), toolName), "\n\n", "\n")
+			} else {
+				// Show full tool call info
+				toolName := messages[i].ToolCall.Name
+				resp[i] = strings.ReplaceAll(fmt.Sprintf("%s\n%s\n[yellow::i][tool call: %s][-:-:-]\nargs: %s\nid: %s\n", icon, messages[i].GetText(), toolName, messages[i].ToolCall.Args, messages[i].ToolCall.ID), "\n\n", "\n")
+			}
 			continue
 		}
-		resp[i] = msg.ToText(i)
+		// Handle tool responses
+		if messages[i].Role == cfg.ToolRole || messages[i].Role == "tool" {
+			// Always show shell commands
+			if messages[i].IsShellCommand {
+				resp[i] = MsgToText(i, &messages[i])
+				continue
+			}
+			// Hide non-shell tool responses when collapsed
+			if toolCollapsed {
+				resp[i] = icon + "\n[yellow::i][tool resp (press Ctrl+T to expand)][-:-:-]\n"
+				continue
+			}
+			// When expanded, show tool responses
+			resp[i] = MsgToText(i, &messages[i])
+			continue
+		}
+		// INFO: skips system msg when showSys is false
+		if !showSys && messages[i].Role == "system" {
+			continue
+		}
+		resp[i] = MsgToText(i, &messages[i])
 	}
 	return resp
 }
@@ -1239,20 +1338,6 @@ func chatToText(messages []models.RoleMsg, showSys bool) string {
 	return text
 }
 
-func removeThinking(chatBody *models.ChatBody) {
-	msgs := []models.RoleMsg{}
-	for _, msg := range chatBody.Messages {
-		// Filter out tool messages and thinking markers
-		if msg.Role == cfg.ToolRole {
-			continue
-		}
-		// find thinking and remove it - use SetText to preserve ContentParts
-		msg.SetText(thinkRE.ReplaceAllString(msg.GetText(), ""))
-		msgs = append(msgs, msg)
-	}
-	chatBody.Messages = msgs
-}
-
 func addNewChat(chatName string) {
 	id, err := store.ChatGetMaxID()
 	if err != nil {
@@ -1307,11 +1392,28 @@ func updateModelLists() {
 	}
 	// if llama.cpp started after gf-lt?
 	localModelsMu.Lock()
-	LocalModels, err = fetchLCPModels()
+	LocalModels, err = fetchLCPModelsWithLoadStatus()
 	localModelsMu.Unlock()
 	if err != nil {
 		logger.Warn("failed to fetch llama.cpp models", "error", err)
 	}
+	// set already loaded model in llama.cpp
+	if strings.Contains(cfg.CurrentAPI, "localhost") || strings.Contains(cfg.CurrentAPI, "127.0.0.1") {
+		localModelsMu.Lock()
+		defer localModelsMu.Unlock()
+		for i := range LocalModels {
+			if strings.Contains(LocalModels[i], models.LoadedMark) {
+				m := strings.TrimPrefix(LocalModels[i], models.LoadedMark)
+				cfg.CurrentModel = m
+				chatBody.Model = m
+				cachedModelColor = "green"
+				updateStatusLine()
+				UpdateToolCapabilities()
+				app.Draw()
+				return
+			}
+		}
+	}
 }
 
 func refreshLocalModelsIfEmpty() {
@@ -1374,15 +1476,6 @@ func init() {
 		os.Exit(1)
 		return
 	}
-	// Set image base directory for path display
-	baseDir := cfg.FilePickerDir
-	if baseDir == "" || baseDir == "." {
-		// Resolve "." to current working directory
-		if wd, err := os.Getwd(); err == nil {
-			baseDir = wd
-		}
-	}
-	models.SetImageBaseDir(baseDir)
 	defaultStarter = []models.RoleMsg{
 		{Role: "system", Content: basicSysMsg},
 		{Role: cfg.AssistantRole, Content: defaultFirstMsg},
@@ -1397,8 +1490,6 @@ func init() {
 	}
 	// load cards
 	basicCard.Role = cfg.AssistantRole
-	// toolCard.Role = cfg.AssistantRole
-	//
 	logLevel.Set(slog.LevelInfo)
 	logger = slog.New(slog.NewTextHandler(logfile, &slog.HandlerOptions{Level: logLevel}))
 	store = storage.NewProviderSQL(cfg.DBPATH, logger)

config.example.toml

@@ -27,7 +27,6 @@ AutoCleanToolCallsFromCtx = false
 RAGEnabled = false
 RAGBatchSize = 1
 RAGWordLimit = 80
-RAGWorkers = 2
 RAGDir = "ragimport"
 # extra tts
 TTS_ENABLED = false

config/config.go

@@ -39,7 +39,6 @@ type Config struct {
 	// rag settings
 	RAGEnabled   bool   `toml:"RAGEnabled"`
 	RAGDir       string `toml:"RAGDir"`
-	RAGWorkers   uint32 `toml:"RAGWorkers"`
 	RAGBatchSize int    `toml:"RAGBatchSize"`
 	RAGWordLimit uint32 `toml:"RAGWordLimit"`
 	// deepseek

docs/config.md

@@ -80,9 +80,6 @@ This document explains how to set up and configure the application using the `co
 #### RAGWordLimit (`80`)
 - Maximum number of words in a batch to tokenize and store.
 
-#### RAGWorkers (`2`)
-- Number of concurrent workers for RAG processing.
-
 #### RAGDir (`"ragimport"`)
 - Directory containing documents for RAG processing.
 

helpfuncs.go

@@ -15,8 +15,6 @@ import (
 	"time"
 	"unicode"
 
-	"math/rand/v2"
-
 	"github.com/rivo/tview"
 )
 
@@ -29,7 +27,6 @@ func startModelColorUpdater() {
 	go func() {
 		ticker := time.NewTicker(5 * time.Second)
 		defer ticker.Stop()
-
 		// Initial check
 		updateCachedModelColor()
 		for range ticker.C {
@@ -44,7 +41,6 @@ func updateCachedModelColor() {
 		cachedModelColor = "orange"
 		return
 	}
-
 	// Check if model is loaded
 	loaded, err := isModelLoaded(chatBody.Model)
 	if err != nil {
@@ -68,6 +64,14 @@ func isASCII(s string) bool {
 	return true
 }
 
+func mapToString[V any](m map[string]V) string {
+	rs := strings.Builder{}
+	for k, v := range m {
+		fmt.Fprintf(&rs, "%v: %v\n", k, v)
+	}
+	return rs.String()
+}
+
 // stripThinkingFromMsg removes thinking blocks from assistant messages.
 // Skips user, tool, and system messages as they may contain thinking examples.
 func stripThinkingFromMsg(msg *models.RoleMsg) *models.RoleMsg {
@@ -375,16 +379,6 @@ func makeStatusLine() string {
 	return statusLine + imageInfo + shellModeInfo
 }
 
-var letters = []rune("abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ")
-
-func randString(n int) string {
-	b := make([]rune, n)
-	for i := range b {
-		b[i] = letters[rand.IntN(len(letters))]
-	}
-	return string(b)
-}
-
 // set of roles within card definition and mention in chat history
 func listChatRoles() []string {
 	currentChat, ok := chatMap[activeChatName]
@@ -465,9 +459,8 @@ func updateFlexLayout() {
 }
 
 func executeCommandAndDisplay(cmdText string) {
-	// Parse the command (split by spaces, but handle quoted arguments)
+	cmdText = strings.TrimSpace(cmdText)
-	cmdParts := parseCommand(cmdText)
+	if cmdText == "" {
-	if len(cmdParts) == 0 {
 		fmt.Fprintf(textView, "\n[red]Error: No command provided[-:-:-]\n")
 		if scrollToEndEnabled {
 			textView.ScrollToEnd()
@@ -475,14 +468,58 @@ func executeCommandAndDisplay(cmdText string) {
 		colorText()
 		return
 	}
-	command := cmdParts[0]
+	workingDir := cfg.FilePickerDir
-	args := []string{}
+	// Handle cd command specially to update working directory
-	if len(cmdParts) > 1 {
+	if strings.HasPrefix(cmdText, "cd ") {
-		args = cmdParts[1:]
+		newDir := strings.TrimPrefix(cmdText, "cd ")
+		newDir = strings.TrimSpace(newDir)
+		// Handle cd ~ or cdHOME
+		if strings.HasPrefix(newDir, "~") {
+			home := os.Getenv("HOME")
+			newDir = strings.Replace(newDir, "~", home, 1)
+		}
+		// Check if directory exists
+		if _, err := os.Stat(newDir); err == nil {
+			workingDir = newDir
+			cfg.FilePickerDir = workingDir
+			// Update shell input label with new directory
+			shellInput.SetLabel(fmt.Sprintf("[%s]$ ", cfg.FilePickerDir))
+			outputContent := workingDir
+			// Add the command being executed to the chat
+			fmt.Fprintf(textView, "\n[-:-:b](%d) <%s>: [-:-:-]\n$ %s\n",
+				len(chatBody.Messages), cfg.ToolRole, cmdText)
+			fmt.Fprintf(textView, "%s\n", outputContent)
+			combinedMsg := models.RoleMsg{
+				Role:    cfg.ToolRole,
+				Content: "$ " + cmdText + "\n\n" + outputContent,
+			}
+			chatBody.Messages = append(chatBody.Messages, combinedMsg)
+			if scrollToEndEnabled {
+				textView.ScrollToEnd()
+			}
+			colorText()
+			return
+		} else {
+			outputContent := "cd: " + newDir + ": No such file or directory"
+			fmt.Fprintf(textView, "\n[-:-:b](%d) <%s>: [-:-:-]\n$ %s\n",
+				len(chatBody.Messages), cfg.ToolRole, cmdText)
+			fmt.Fprintf(textView, "[red]%s[-:-:-]\n", outputContent)
+			combinedMsg := models.RoleMsg{
+				Role:    cfg.ToolRole,
+				Content: "$ " + cmdText + "\n\n" + outputContent,
+			}
+			chatBody.Messages = append(chatBody.Messages, combinedMsg)
+			if scrollToEndEnabled {
+				textView.ScrollToEnd()
+			}
+			colorText()
+			return
+		}
 	}
-	// Create the command execution
+
-	cmd := exec.Command(command, args...)
+	// Use /bin/sh to support pipes, redirects, etc.
-	cmd.Dir = cfg.FilePickerDir
+	cmd := exec.Command("/bin/sh", "-c", cmdText)
+	cmd.Dir = workingDir
 	// Execute the command and get output
 	output, err := cmd.CombinedOutput()
 	// Add the command being executed to the chat
@@ -531,42 +568,6 @@ func executeCommandAndDisplay(cmdText string) {
 	shellHistoryPos = -1
 }
 
-// parseCommand splits command string handling quotes properly
-func parseCommand(cmd string) []string {
-	var args []string
-	var current string
-	var inQuotes bool
-	var quoteChar rune
-	for _, r := range cmd {
-		switch r {
-		case '"', '\'':
-			if inQuotes {
-				if r == quoteChar {
-					inQuotes = false
-				} else {
-					current += string(r)
-				}
-			} else {
-				inQuotes = true
-				quoteChar = r
-			}
-		case ' ', '\t':
-			if inQuotes {
-				current += string(r)
-			} else if current != "" {
-				args = append(args, current)
-				current = ""
-			}
-		default:
-			current += string(r)
-		}
-	}
-	if current != "" {
-		args = append(args, current)
-	}
-	return args
-}
-
 // == search ==
 
 // Global variables for search state
@@ -806,3 +807,91 @@ func scanFiles(dir, filter string) []string {
 	scanRecursive(dir, 0, "")
 	return files
 }
+
+// models logic that is too complex for models package
+func MsgToText(i int, m *models.RoleMsg) string {
+	var contentStr string
+	var imageIndicators []string
+	if !m.HasContentParts {
+		contentStr = m.Content
+	} else {
+		var textParts []string
+		for _, part := range m.ContentParts {
+			switch p := part.(type) {
+			case models.TextContentPart:
+				if p.Type == "text" {
+					textParts = append(textParts, p.Text)
+				}
+			case models.ImageContentPart:
+				displayPath := p.Path
+				if displayPath == "" {
+					displayPath = "image"
+				} else {
+					displayPath = extractDisplayPath(displayPath, cfg.FilePickerDir)
+				}
+				imageIndicators = append(imageIndicators, fmt.Sprintf("[orange::i][image: %s][-:-:-]", displayPath))
+			case map[string]any:
+				if partType, exists := p["type"]; exists {
+					switch partType {
+					case "text":
+						if textVal, textExists := p["text"]; textExists {
+							if textStr, isStr := textVal.(string); isStr {
+								textParts = append(textParts, textStr)
+							}
+						}
+					case "image_url":
+						var displayPath string
+						if pathVal, pathExists := p["path"]; pathExists {
+							if pathStr, isStr := pathVal.(string); isStr && pathStr != "" {
+								displayPath = extractDisplayPath(pathStr, cfg.FilePickerDir)
+							}
+						}
+						if displayPath == "" {
+							displayPath = "image"
+						}
+						imageIndicators = append(imageIndicators, fmt.Sprintf("[orange::i][image: %s][-:-:-]", displayPath))
+					}
+				}
+			}
+		}
+		contentStr = strings.Join(textParts, " ") + " "
+	}
+	contentStr, _ = strings.CutPrefix(contentStr, m.Role+":")
+	icon := fmt.Sprintf("(%d) <%s>: ", i, m.Role)
+	var finalContent strings.Builder
+	if len(imageIndicators) > 0 {
+		for _, indicator := range imageIndicators {
+			finalContent.WriteString(indicator)
+			finalContent.WriteString("\n")
+		}
+	}
+	finalContent.WriteString(contentStr)
+	if m.Stats != nil {
+		fmt.Fprintf(&finalContent, "\n[gray::i][%d tok, %.1fs, %.1f t/s][-:-:-]", m.Stats.Tokens, m.Stats.Duration, m.Stats.TokensPerSec)
+	}
+	textMsg := fmt.Sprintf("[-:-:b]%s[-:-:-]\n%s\n", icon, finalContent.String())
+	return strings.ReplaceAll(textMsg, "\n\n", "\n")
+}
+
+// extractDisplayPath returns a path suitable for display, potentially relative to imageBaseDir
+func extractDisplayPath(p, bp string) string {
+	if p == "" {
+		return ""
+	}
+	// If base directory is set, try to make path relative to it
+	if bp != "" {
+		if rel, err := filepath.Rel(bp, p); err == nil {
+			// Check if relative path doesn't start with ".." (meaning it's within base dir)
+			// If it starts with "..", we might still want to show it as relative
+			// but for now we show full path if it goes outside base dir
+			if !strings.HasPrefix(rel, "..") {
+				p = rel
+			}
+		}
+	}
+	// Truncate long paths to last 60 characters if needed
+	if len(p) > 60 {
+		return "..." + p[len(p)-60:]
+	}
+	return p
+}

llm.go

@@ -3,7 +3,6 @@ package main
 import (
 	"bytes"
 	"encoding/json"
-	"errors"
 	"gf-lt/models"
 	"io"
 	"strings"
@@ -14,8 +13,8 @@ var lastImg string             // for ctrl+j
 
 // containsToolSysMsg checks if the toolSysMsg already exists in the chat body
 func containsToolSysMsg() bool {
-	for _, msg := range chatBody.Messages {
+	for i := range chatBody.Messages {
-		if msg.Role == cfg.ToolRole && msg.Content == toolSysMsg {
+		if chatBody.Messages[i].Role == cfg.ToolRole && chatBody.Messages[i].Content == toolSysMsg {
 			return true
 		}
 	}
@@ -119,25 +118,22 @@ func (lcp LCPCompletion) FormMsg(msg, role string, resume bool) (io.Reader, erro
 	logger.Debug("formmsg lcpcompletion", "link", cfg.CurrentAPI)
 	localImageAttachmentPath := imageAttachmentPath
 	var multimodalData []string
-	if localImageAttachmentPath != "" {
-		imageURL, err := models.CreateImageURLFromPath(localImageAttachmentPath)
-		if err != nil {
-			logger.Error("failed to create image URL from path for completion",
-				"error", err, "path", localImageAttachmentPath)
-			return nil, err
-		}
-		// Extract base64 part from data URL (e.g., "data:image/jpeg;base64,...")
-		parts := strings.SplitN(imageURL, ",", 2)
-		if len(parts) == 2 {
-			multimodalData = append(multimodalData, parts[1])
-		} else {
-			logger.Error("invalid image data URL format", "url", imageURL)
-			return nil, errors.New("invalid image data URL format")
-		}
-		imageAttachmentPath = "" // Clear the attachment after use
-	}
 	if msg != "" { // otherwise let the bot to continue
-		newMsg := models.RoleMsg{Role: role, Content: msg}
+		var newMsg models.RoleMsg
+		if localImageAttachmentPath != "" {
+			newMsg = models.NewMultimodalMsg(role, []any{})
+			newMsg.AddTextPart(msg)
+			imageURL, err := models.CreateImageURLFromPath(localImageAttachmentPath)
+			if err != nil {
+				logger.Error("failed to create image URL from path for completion",
+					"error", err, "path", localImageAttachmentPath)
+				return nil, err
+			}
+			newMsg.AddImagePart(imageURL, localImageAttachmentPath)
+			imageAttachmentPath = "" // Clear the attachment after use
+		} else { // not a multimodal msg or image passed in tool call
+			newMsg = models.RoleMsg{Role: role, Content: msg}
+		}
 		newMsg = *processMessageTag(&newMsg)
 		chatBody.Messages = append(chatBody.Messages, newMsg)
 	}
@@ -146,22 +142,40 @@ func (lcp LCPCompletion) FormMsg(msg, role string, resume bool) (io.Reader, erro
 		chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
 	}
 	filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages)
+	// Build prompt and extract images inline as we process each message
 	messages := make([]string, len(filteredMessages))
-	for i, m := range filteredMessages {
+	for i := range filteredMessages {
-		messages[i] = stripThinkingFromMsg(&m).ToPrompt()
+		m := stripThinkingFromMsg(&filteredMessages[i])
+		messages[i] = m.ToPrompt()
+		// Extract images from this message and add marker inline
+		if len(m.ContentParts) > 0 {
+			for _, part := range m.ContentParts {
+				var imgURL string
+				// Check for struct type
+				if imgPart, ok := part.(models.ImageContentPart); ok {
+					imgURL = imgPart.ImageURL.URL
+				} else if partMap, ok := part.(map[string]any); ok {
+					// Check for map type (from JSON unmarshaling)
+					if partType, exists := partMap["type"]; exists && partType == "image_url" {
+						if imgURLMap, ok := partMap["image_url"].(map[string]any); ok {
+							if url, ok := imgURLMap["url"].(string); ok {
+								imgURL = url
+							}
+						}
+					}
+				}
+				if imgURL != "" {
+					// Extract base64 part from data URL (e.g., "data:image/jpeg;base64,...")
+					parts := strings.SplitN(imgURL, ",", 2)
+					if len(parts) == 2 {
+						multimodalData = append(multimodalData, parts[1])
+						messages[i] += " <__media__>"
+					}
+				}
+			}
+		}
 	}
 	prompt := strings.Join(messages, "\n")
-	// Add multimodal media markers to the prompt text when multimodal data is present
-	// This is required by llama.cpp multimodal models so they know where to insert media
-	if len(multimodalData) > 0 {
-		// Add a media marker for each item in the multimodal data
-		var sb strings.Builder
-		sb.WriteString(prompt)
-		for range multimodalData {
-			sb.WriteString(" <__media__>") // llama.cpp default multimodal marker
-		}
-		prompt = sb.String()
-	}
 	// needs to be after <__media__> if there are images
 	if !resume {
 		botMsgStart := "\n" + botPersona + ":\n"
@@ -289,14 +303,23 @@ func (op LCPChat) FormMsg(msg, role string, resume bool) (io.Reader, error) {
 		Model:    chatBody.Model,
 		Stream:   chatBody.Stream,
 	}
-	for i, msg := range filteredMessages {
+	for i := range filteredMessages {
-		strippedMsg := *stripThinkingFromMsg(&msg)
+		strippedMsg := *stripThinkingFromMsg(&filteredMessages[i])
-		if strippedMsg.Role == cfg.UserRole {
+		switch strippedMsg.Role {
+		case cfg.UserRole:
 			bodyCopy.Messages[i] = strippedMsg
 			bodyCopy.Messages[i].Role = "user"
-		} else {
+		case cfg.AssistantRole:
+			bodyCopy.Messages[i] = strippedMsg
+			bodyCopy.Messages[i].Role = "assistant"
+		case cfg.ToolRole:
+			bodyCopy.Messages[i] = strippedMsg
+			bodyCopy.Messages[i].Role = "tool"
+		default:
 			bodyCopy.Messages[i] = strippedMsg
 		}
+		// Clear ToolCalls - they're stored in chat history for display but not sent to LLM
+		// bodyCopy.Messages[i].ToolCall = nil
 	}
 	// Clean null/empty messages to prevent API issues
 	bodyCopy.Messages = consolidateAssistantMessages(bodyCopy.Messages)
@@ -358,8 +381,8 @@ func (ds DeepSeekerCompletion) FormMsg(msg, role string, resume bool) (io.Reader
 	}
 	filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages)
 	messages := make([]string, len(filteredMessages))
-	for i, m := range filteredMessages {
+	for i := range filteredMessages {
-		messages[i] = stripThinkingFromMsg(&m).ToPrompt()
+		messages[i] = stripThinkingFromMsg(&filteredMessages[i]).ToPrompt()
 	}
 	prompt := strings.Join(messages, "\n")
 	// strings builder?
@@ -429,14 +452,27 @@ func (ds DeepSeekerChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
 		Model:    chatBody.Model,
 		Stream:   chatBody.Stream,
 	}
-	for i, msg := range filteredMessages {
+	for i := range filteredMessages {
-		strippedMsg := *stripThinkingFromMsg(&msg)
+		strippedMsg := *stripThinkingFromMsg(&filteredMessages[i])
-		if strippedMsg.Role == cfg.UserRole || i == 1 {
+		switch strippedMsg.Role {
+		case cfg.UserRole:
+			if i == 1 {
+				bodyCopy.Messages[i] = strippedMsg
+				bodyCopy.Messages[i].Role = "user"
+			} else {
+				bodyCopy.Messages[i] = strippedMsg
+			}
+		case cfg.AssistantRole:
 			bodyCopy.Messages[i] = strippedMsg
-			bodyCopy.Messages[i].Role = "user"
+			bodyCopy.Messages[i].Role = "assistant"
-		} else {
+		case cfg.ToolRole:
+			bodyCopy.Messages[i] = strippedMsg
+			bodyCopy.Messages[i].Role = "tool"
+		default:
 			bodyCopy.Messages[i] = strippedMsg
 		}
+		// Clear ToolCalls - they're stored in chat history for display but not sent to LLM
+		// bodyCopy.Messages[i].ToolCall = nil
 	}
 	// Clean null/empty messages to prevent API issues
 	bodyCopy.Messages = consolidateAssistantMessages(bodyCopy.Messages)
@@ -489,8 +525,8 @@ func (or OpenRouterCompletion) FormMsg(msg, role string, resume bool) (io.Reader
 	}
 	filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages)
 	messages := make([]string, len(filteredMessages))
-	for i, m := range filteredMessages {
+	for i := range filteredMessages {
-		messages[i] = stripThinkingFromMsg(&m).ToPrompt()
+		messages[i] = stripThinkingFromMsg(&filteredMessages[i]).ToPrompt()
 	}
 	prompt := strings.Join(messages, "\n")
 	// strings builder?
@@ -593,14 +629,24 @@ func (or OpenRouterChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
 		Model:    chatBody.Model,
 		Stream:   chatBody.Stream,
 	}
-	for i, msg := range filteredMessages {
+	for i := range filteredMessages {
-		strippedMsg := *stripThinkingFromMsg(&msg)
+		strippedMsg := *stripThinkingFromMsg(&filteredMessages[i])
-		bodyCopy.Messages[i] = strippedMsg
+		switch strippedMsg.Role {
-		// Standardize role if it's a user role
+		case cfg.UserRole:
-		if bodyCopy.Messages[i].Role == cfg.UserRole {
 			bodyCopy.Messages[i] = strippedMsg
 			bodyCopy.Messages[i].Role = "user"
+		case cfg.AssistantRole:
+			bodyCopy.Messages[i] = strippedMsg
+			bodyCopy.Messages[i].Role = "assistant"
+		case cfg.ToolRole:
+			bodyCopy.Messages[i] = strippedMsg
+			bodyCopy.Messages[i].Role = "tool"
+		default:
+			bodyCopy.Messages[i] = strippedMsg
 		}
+		// Clear ToolCalls - they're stored in chat history for display but not sent to LLM
+		// literally deletes data that we need
+		// bodyCopy.Messages[i].ToolCall = nil
 	}
 	// Clean null/empty messages to prevent API issues
 	bodyCopy.Messages = consolidateAssistantMessages(bodyCopy.Messages)

main.go

@@ -16,6 +16,7 @@ var (
 	shellHistory      []string
 	shellHistoryPos   int = -1
 	thinkingCollapsed     = false
+	toolCollapsed         = true
 	statusLineTempl       = "help (F12) | chat: [orange:-:b]%s[-:-:-] (F1) | [%s:-:b]tool use[-:-:-] (ctrl+k) | model: [%s:-:b]%s[-:-:-] (ctrl+l) | [%s:-:b]skip LLM resp[-:-:-] (F10)\nAPI: [orange:-:b]%s[-:-:-] (ctrl+v) | writing as: [orange:-:b]%s[-:-:-] (ctrl+q) | bot will write as [orange:-:b]%s[-:-:-] (ctrl+x)"
 	focusSwitcher         = map[tview.Primitive]tview.Primitive{}
 )

main_test.go

@@ -1,42 +0,0 @@
-package main
-
-import (
-	"fmt"
-	"gf-lt/config"
-	"gf-lt/models"
-	"strings"
-	"testing"
-)
-
-func TestRemoveThinking(t *testing.T) {
-	cases := []struct {
-		cb       *models.ChatBody
-		toolMsgs uint8
-	}{
-		{cb: &models.ChatBody{
-			Stream: true,
-			Messages: []models.RoleMsg{
-				{Role: "tool", Content: "should be ommited"},
-				{Role: "system", Content: "should stay"},
-				{Role: "user", Content: "hello, how are you?"},
-				{Role: "assistant", Content: "Oh, hi. <think>I should thank user and continue the conversation</think> I am geat, thank you! How are you?"},
-			},
-		},
-			toolMsgs: uint8(1),
-		},
-	}
-	for i, tc := range cases {
-					t.Run(fmt.Sprintf("run_%d", i), func(t *testing.T) {
-						cfg = &config.Config{ToolRole: "tool"} // Initialize cfg.ToolRole for test
-						mNum := len(tc.cb.Messages)
-						removeThinking(tc.cb)
-						if len(tc.cb.Messages) != mNum-int(tc.toolMsgs) {
-							t.Errorf("failed to delete tools msg %v; expected %d, got %d", tc.cb.Messages, mNum-int(tc.toolMsgs), len(tc.cb.Messages))
-						}
-						for _, msg := range tc.cb.Messages {
-							if strings.Contains(msg.Content, "<think>") {
-								t.Errorf("msg contains think tag; msg: %s\n", msg.Content)
-							}
-						}
-					})	}
-}

models/consts.go

@@ -0,0 +1,13 @@
+package models
+
+const (
+	LoadedMark        = "(loaded) "
+	ToolRespMultyType = "multimodel_content"
+)
+
+type APIType int
+
+const (
+	APITypeChat APIType = iota
+	APITypeCompletion
+)

models/models.go

@@ -5,28 +5,21 @@ import (
 	"encoding/json"
 	"fmt"
 	"os"
-	"path/filepath"
 	"strings"
 )
 
-var (
-	// imageBaseDir is the base directory for displaying image paths.
-	// If set, image paths will be shown relative to this directory.
-	imageBaseDir = ""
-)
-
-// SetImageBaseDir sets the base directory for displaying image paths.
-// If dir is empty, full paths will be shown.
-func SetImageBaseDir(dir string) {
-	imageBaseDir = dir
-}
-
 type FuncCall struct {
 	ID   string            `json:"id,omitempty"`
 	Name string            `json:"name"`
 	Args map[string]string `json:"args"`
 }
 
+type ToolCall struct {
+	ID   string `json:"id,omitempty"`
+	Name string `json:"name"`
+	Args string `json:"arguments"`
+}
+
 type LLMResp struct {
 	Choices []struct {
 		FinishReason string `json:"finish_reason"`
@@ -108,46 +101,56 @@ type RoleMsg struct {
 	Role            string         `json:"role"`
 	Content         string         `json:"-"`
 	ContentParts    []any          `json:"-"`
-	ToolCallID      string         `json:"tool_call_id,omitempty"` // For tool response messages
+	ToolCallID      string         `json:"tool_call_id,omitempty"`     // For tool response messages
+	ToolCall        *ToolCall      `json:"tool_call,omitempty"`        // For assistant messages with tool calls
+	IsShellCommand  bool           `json:"is_shell_command,omitempty"` // True for shell command outputs (always shown)
 	KnownTo         []string       `json:"known_to,omitempty"`
 	Stats           *ResponseStats `json:"stats"`
-	hasContentParts bool           // Flag to indicate which content type to marshal
+	HasContentParts bool           // Flag to indicate which content type to marshal
 }
 
 // MarshalJSON implements custom JSON marshaling for RoleMsg
 //
 //nolint:gocritic
 func (m RoleMsg) MarshalJSON() ([]byte, error) {
-	if m.hasContentParts {
+	if m.HasContentParts {
 		// Use structured content format
 		aux := struct {
-			Role       string         `json:"role"`
+			Role           string         `json:"role"`
-			Content    []any          `json:"content"`
+			Content        []any          `json:"content"`
-			ToolCallID string         `json:"tool_call_id,omitempty"`
+			ToolCallID     string         `json:"tool_call_id,omitempty"`
-			KnownTo    []string       `json:"known_to,omitempty"`
+			ToolCall       *ToolCall      `json:"tool_call,omitempty"`
-			Stats      *ResponseStats `json:"stats,omitempty"`
+			IsShellCommand bool           `json:"is_shell_command,omitempty"`
+			KnownTo        []string       `json:"known_to,omitempty"`
+			Stats          *ResponseStats `json:"stats,omitempty"`
 		}{
-			Role:       m.Role,
+			Role:           m.Role,
-			Content:    m.ContentParts,
+			Content:        m.ContentParts,
-			ToolCallID: m.ToolCallID,
+			ToolCallID:     m.ToolCallID,
-			KnownTo:    m.KnownTo,
+			ToolCall:       m.ToolCall,
-			Stats:      m.Stats,
+			IsShellCommand: m.IsShellCommand,
+			KnownTo:        m.KnownTo,
+			Stats:          m.Stats,
 		}
 		return json.Marshal(aux)
 	} else {
 		// Use simple content format
 		aux := struct {
-			Role       string         `json:"role"`
+			Role           string         `json:"role"`
-			Content    string         `json:"content"`
+			Content        string         `json:"content"`
-			ToolCallID string         `json:"tool_call_id,omitempty"`
+			ToolCallID     string         `json:"tool_call_id,omitempty"`
-			KnownTo    []string       `json:"known_to,omitempty"`
+			ToolCall       *ToolCall      `json:"tool_call,omitempty"`
-			Stats      *ResponseStats `json:"stats,omitempty"`
+			IsShellCommand bool           `json:"is_shell_command,omitempty"`
+			KnownTo        []string       `json:"known_to,omitempty"`
+			Stats          *ResponseStats `json:"stats,omitempty"`
 		}{
-			Role:       m.Role,
+			Role:           m.Role,
-			Content:    m.Content,
+			Content:        m.Content,
-			ToolCallID: m.ToolCallID,
+			ToolCallID:     m.ToolCallID,
-			KnownTo:    m.KnownTo,
+			ToolCall:       m.ToolCall,
-			Stats:      m.Stats,
+			IsShellCommand: m.IsShellCommand,
+			KnownTo:        m.KnownTo,
+			Stats:          m.Stats,
 		}
 		return json.Marshal(aux)
 	}
@@ -157,29 +160,35 @@ func (m RoleMsg) MarshalJSON() ([]byte, error) {
 func (m *RoleMsg) UnmarshalJSON(data []byte) error {
 	// First, try to unmarshal as structured content format
 	var structured struct {
-		Role       string         `json:"role"`
+		Role           string         `json:"role"`
-		Content    []any          `json:"content"`
+		Content        []any          `json:"content"`
-		ToolCallID string         `json:"tool_call_id,omitempty"`
+		ToolCallID     string         `json:"tool_call_id,omitempty"`
-		KnownTo    []string       `json:"known_to,omitempty"`
+		ToolCall       *ToolCall      `json:"tool_call,omitempty"`
-		Stats      *ResponseStats `json:"stats,omitempty"`
+		IsShellCommand bool           `json:"is_shell_command,omitempty"`
+		KnownTo        []string       `json:"known_to,omitempty"`
+		Stats          *ResponseStats `json:"stats,omitempty"`
 	}
 	if err := json.Unmarshal(data, &structured); err == nil && len(structured.Content) > 0 {
 		m.Role = structured.Role
 		m.ContentParts = structured.Content
 		m.ToolCallID = structured.ToolCallID
+		m.ToolCall = structured.ToolCall
+		m.IsShellCommand = structured.IsShellCommand
 		m.KnownTo = structured.KnownTo
 		m.Stats = structured.Stats
-		m.hasContentParts = true
+		m.HasContentParts = true
 		return nil
 	}
 
 	// Otherwise, unmarshal as simple content format
 	var simple struct {
-		Role       string         `json:"role"`
+		Role           string         `json:"role"`
-		Content    string         `json:"content"`
+		Content        string         `json:"content"`
-		ToolCallID string         `json:"tool_call_id,omitempty"`
+		ToolCallID     string         `json:"tool_call_id,omitempty"`
-		KnownTo    []string       `json:"known_to,omitempty"`
+		ToolCall       *ToolCall      `json:"tool_call,omitempty"`
-		Stats      *ResponseStats `json:"stats,omitempty"`
+		IsShellCommand bool           `json:"is_shell_command,omitempty"`
+		KnownTo        []string       `json:"known_to,omitempty"`
+		Stats          *ResponseStats `json:"stats,omitempty"`
 	}
 	if err := json.Unmarshal(data, &simple); err != nil {
 		return err
@@ -187,79 +196,17 @@ func (m *RoleMsg) UnmarshalJSON(data []byte) error {
 	m.Role = simple.Role
 	m.Content = simple.Content
 	m.ToolCallID = simple.ToolCallID
+	m.ToolCall = simple.ToolCall
+	m.IsShellCommand = simple.IsShellCommand
 	m.KnownTo = simple.KnownTo
 	m.Stats = simple.Stats
-	m.hasContentParts = false
+	m.HasContentParts = false
 	return nil
 }
 
-func (m *RoleMsg) ToText(i int) string {
-	var contentStr string
-	var imageIndicators []string
-	if !m.hasContentParts {
-		contentStr = m.Content
-	} else {
-		var textParts []string
-		for _, part := range m.ContentParts {
-			switch p := part.(type) {
-			case TextContentPart:
-				if p.Type == "text" {
-					textParts = append(textParts, p.Text)
-				}
-			case ImageContentPart:
-				displayPath := p.Path
-				if displayPath == "" {
-					displayPath = "image"
-				} else {
-					displayPath = extractDisplayPath(displayPath)
-				}
-				imageIndicators = append(imageIndicators, fmt.Sprintf("[orange::i][image: %s][-:-:-]", displayPath))
-			case map[string]any:
-				if partType, exists := p["type"]; exists {
-					switch partType {
-					case "text":
-						if textVal, textExists := p["text"]; textExists {
-							if textStr, isStr := textVal.(string); isStr {
-								textParts = append(textParts, textStr)
-							}
-						}
-					case "image_url":
-						var displayPath string
-						if pathVal, pathExists := p["path"]; pathExists {
-							if pathStr, isStr := pathVal.(string); isStr && pathStr != "" {
-								displayPath = extractDisplayPath(pathStr)
-							}
-						}
-						if displayPath == "" {
-							displayPath = "image"
-						}
-						imageIndicators = append(imageIndicators, fmt.Sprintf("[orange::i][image: %s][-:-:-]", displayPath))
-					}
-				}
-			}
-		}
-		contentStr = strings.Join(textParts, " ") + " "
-	}
-	contentStr, _ = strings.CutPrefix(contentStr, m.Role+":")
-	icon := fmt.Sprintf("(%d) <%s>: ", i, m.Role)
-	var finalContent strings.Builder
-	if len(imageIndicators) > 0 {
-		for _, indicator := range imageIndicators {
-			finalContent.WriteString(indicator)
-			finalContent.WriteString("\n")
-		}
-	}
-	finalContent.WriteString(contentStr)
-	if m.Stats != nil {
-		fmt.Fprintf(&finalContent, "\n[gray::i][%d tok, %.1fs, %.1f t/s][-:-:-]", m.Stats.Tokens, m.Stats.Duration, m.Stats.TokensPerSec)
-	}
-	textMsg := fmt.Sprintf("[-:-:b]%s[-:-:-]\n%s\n", icon, finalContent.String())
-	return strings.ReplaceAll(textMsg, "\n\n", "\n")
-}
-
 func (m *RoleMsg) ToPrompt() string {
 	var contentStr string
-	if !m.hasContentParts {
+	if !m.HasContentParts {
 		contentStr = m.Content
 	} else {
 		// For structured content, just take the text parts
@@ -292,7 +239,7 @@ func NewRoleMsg(role, content string) RoleMsg {
 	return RoleMsg{
 		Role:            role,
 		Content:         content,
-		hasContentParts: false,
+		HasContentParts: false,
 	}
 }
 
@@ -301,7 +248,7 @@ func NewMultimodalMsg(role string, contentParts []any) RoleMsg {
 	return RoleMsg{
 		Role:            role,
 		ContentParts:    contentParts,
-		hasContentParts: true,
+		HasContentParts: true,
 	}
 }
 
@@ -310,7 +257,7 @@ func (m *RoleMsg) HasContent() bool {
 	if m.Content != "" {
 		return true
 	}
-	if m.hasContentParts && len(m.ContentParts) > 0 {
+	if m.HasContentParts && len(m.ContentParts) > 0 {
 		return true
 	}
 	return false
@@ -318,7 +265,7 @@ func (m *RoleMsg) HasContent() bool {
 
 // IsContentParts returns true if the message uses structured content parts
 func (m *RoleMsg) IsContentParts() bool {
-	return m.hasContentParts
+	return m.HasContentParts
 }
 
 // GetContentParts returns the content parts of the message
@@ -335,14 +282,16 @@ func (m *RoleMsg) Copy() RoleMsg {
 		ToolCallID:      m.ToolCallID,
 		KnownTo:         m.KnownTo,
 		Stats:           m.Stats,
-		hasContentParts: m.hasContentParts,
+		HasContentParts: m.HasContentParts,
+		ToolCall:        m.ToolCall,
+		IsShellCommand:  m.IsShellCommand,
 	}
 }
 
 // GetText returns the text content of the message, handling both
 // simple Content and multimodal ContentParts formats.
 func (m *RoleMsg) GetText() string {
-	if !m.hasContentParts {
+	if !m.HasContentParts {
 		return m.Content
 	}
 	var textParts []string
@@ -371,7 +320,7 @@ func (m *RoleMsg) GetText() string {
 // ContentParts (multimodal), it updates the text parts while preserving
 // images. If not, it sets the simple Content field.
 func (m *RoleMsg) SetText(text string) {
-	if !m.hasContentParts {
+	if !m.HasContentParts {
 		m.Content = text
 		return
 	}
@@ -401,14 +350,14 @@ func (m *RoleMsg) SetText(text string) {
 
 // AddTextPart adds a text content part to the message
 func (m *RoleMsg) AddTextPart(text string) {
-	if !m.hasContentParts {
+	if !m.HasContentParts {
 		// Convert to content parts format
 		if m.Content != "" {
 			m.ContentParts = []any{TextContentPart{Type: "text", Text: m.Content}}
 		} else {
 			m.ContentParts = []any{}
 		}
-		m.hasContentParts = true
+		m.HasContentParts = true
 	}
 	textPart := TextContentPart{Type: "text", Text: text}
 	m.ContentParts = append(m.ContentParts, textPart)
@@ -416,14 +365,14 @@ func (m *RoleMsg) AddTextPart(text string) {
 
 // AddImagePart adds an image content part to the message
 func (m *RoleMsg) AddImagePart(imageURL, imagePath string) {
-	if !m.hasContentParts {
+	if !m.HasContentParts {
 		// Convert to content parts format
 		if m.Content != "" {
 			m.ContentParts = []any{TextContentPart{Type: "text", Text: m.Content}}
 		} else {
 			m.ContentParts = []any{}
 		}
-		m.hasContentParts = true
+		m.HasContentParts = true
 	}
 	imagePart := ImageContentPart{
 		Type: "image_url",
@@ -442,7 +391,6 @@ func CreateImageURLFromPath(imagePath string) (string, error) {
 	if err != nil {
 		return "", err
 	}
-
 	// Determine the image format based on file extension
 	var mimeType string
 	switch {
@@ -459,39 +407,12 @@ func CreateImageURLFromPath(imagePath string) (string, error) {
 	default:
 		mimeType = "image/jpeg" // default
 	}
-
 	// Encode to base64
 	encoded := base64.StdEncoding.EncodeToString(data)
-
 	// Create data URL
 	return fmt.Sprintf("data:%s;base64,%s", mimeType, encoded), nil
 }
 
-// extractDisplayPath returns a path suitable for display, potentially relative to imageBaseDir
-func extractDisplayPath(p string) string {
-	if p == "" {
-		return ""
-	}
-
-	// If base directory is set, try to make path relative to it
-	if imageBaseDir != "" {
-		if rel, err := filepath.Rel(imageBaseDir, p); err == nil {
-			// Check if relative path doesn't start with ".." (meaning it's within base dir)
-			// If it starts with "..", we might still want to show it as relative
-			// but for now we show full path if it goes outside base dir
-			if !strings.HasPrefix(rel, "..") {
-				p = rel
-			}
-		}
-	}
-
-	// Truncate long paths to last 60 characters if needed
-	if len(p) > 60 {
-		return "..." + p[len(p)-60:]
-	}
-	return p
-}
-
 type ChatBody struct {
 	Model    string    `json:"model"`
 	Stream   bool      `json:"stream"`
@@ -499,16 +420,16 @@ type ChatBody struct {
 }
 
 func (cb *ChatBody) Rename(oldname, newname string) {
-	for i, m := range cb.Messages {
+	for i := range cb.Messages {
-		cb.Messages[i].Content = strings.ReplaceAll(m.Content, oldname, newname)
+		cb.Messages[i].Content = strings.ReplaceAll(cb.Messages[i].Content, oldname, newname)
-		cb.Messages[i].Role = strings.ReplaceAll(m.Role, oldname, newname)
+		cb.Messages[i].Role = strings.ReplaceAll(cb.Messages[i].Role, oldname, newname)
 	}
 }
 
 func (cb *ChatBody) ListRoles() []string {
 	namesMap := make(map[string]struct{})
-	for _, m := range cb.Messages {
+	for i := range cb.Messages {
-		namesMap[m.Role] = struct{}{}
+		namesMap[cb.Messages[i].Role] = struct{}{}
 	}
 	resp := make([]string, len(namesMap))
 	i := 0
@@ -595,24 +516,6 @@ type OpenAIReq struct {
 
 // ===
 
-// type LLMModels struct {
-// 	Object string `json:"object"`
-// 	Data   []struct {
-// 		ID      string `json:"id"`
-// 		Object  string `json:"object"`
-// 		Created int    `json:"created"`
-// 		OwnedBy string `json:"owned_by"`
-// 		Meta    struct {
-// 			VocabType int   `json:"vocab_type"`
-// 			NVocab    int   `json:"n_vocab"`
-// 			NCtxTrain int   `json:"n_ctx_train"`
-// 			NEmbd     int   `json:"n_embd"`
-// 			NParams   int64 `json:"n_params"`
-// 			Size      int64 `json:"size"`
-// 		} `json:"meta"`
-// 	} `json:"data"`
-// }
-
 type LlamaCPPReq struct {
 	Model  string `json:"model"`
 	Stream bool   `json:"stream"`
@@ -705,6 +608,20 @@ func (lcp *LCPModels) ListModels() []string {
 	return resp
 }
 
+func (lcp *LCPModels) HasVision(modelID string) bool {
+	for _, m := range lcp.Data {
+		if m.ID == modelID {
+			args := m.Status.Args
+			for i := 0; i < len(args)-1; i++ {
+				if args[i] == "--mmproj" {
+					return true
+				}
+			}
+		}
+	}
+	return false
+}
+
 type ResponseStats struct {
 	Tokens       int
 	Duration     float64
@@ -718,9 +635,7 @@ type ChatRoundReq struct {
 	Resume  bool
 }
 
-type APIType int
+type MultimodalToolResp struct {
-
+	Type  string              `json:"type"`
-const (
+	Parts []map[string]string `json:"parts"`
-	APITypeChat APIType = iota
+}
-	APITypeCompletion
-)

models/models_test.go

@@ -1,161 +0,0 @@
-package models
-import (
-	"strings"
-	"testing"
-)
-func TestRoleMsgToTextWithImages(t *testing.T) {
-	tests := []struct {
-		name     string
-		msg      RoleMsg
-		index    int
-		expected string // substring to check
-	}{
-		{
-			name:  "text and image",
-			index: 0,
-			msg: func() RoleMsg {
-				msg := NewMultimodalMsg("user", []interface{}{})
-				msg.AddTextPart("Look at this picture")
-				msg.AddImagePart("data:image/jpeg;base64,abc123", "/home/user/Pictures/cat.jpg")
-				return msg
-			}(),
-			expected: "[orange::i][image: /home/user/Pictures/cat.jpg][-:-:-]",
-		},
-		{
-			name:  "image only",
-			index: 1,
-			msg: func() RoleMsg {
-				msg := NewMultimodalMsg("user", []interface{}{})
-				msg.AddImagePart("data:image/png;base64,xyz789", "/tmp/screenshot_20250217_123456.png")
-				return msg
-			}(),
-			expected: "[orange::i][image: /tmp/screenshot_20250217_123456.png][-:-:-]",
-		},
-		{
-			name:  "long filename truncated",
-			index: 2,
-			msg: func() RoleMsg {
-				msg := NewMultimodalMsg("user", []interface{}{})
-				msg.AddTextPart("Check this")
-				msg.AddImagePart("data:image/jpeg;base64,foo", "/very/long/path/to/a/really_long_filename_that_exceeds_forty_characters.jpg")
-				return msg
-			}(),
-			expected: "[orange::i][image: .../to/a/really_long_filename_that_exceeds_forty_characters.jpg][-:-:-]",
-		},
-		{
-			name:  "multiple images",
-			index: 3,
-			msg: func() RoleMsg {
-				msg := NewMultimodalMsg("user", []interface{}{})
-				msg.AddTextPart("Multiple images")
-				msg.AddImagePart("data:image/jpeg;base64,a", "/path/img1.jpg")
-				msg.AddImagePart("data:image/png;base64,b", "/path/img2.png")
-				return msg
-			}(),
-			expected: "[orange::i][image: /path/img1.jpg][-:-:-]\n[orange::i][image: /path/img2.png][-:-:-]",
-		},
-		{
-			name:  "old format without path",
-			index: 4,
-			msg: RoleMsg{
-				Role:            "user",
-				hasContentParts: true,
-				ContentParts: []interface{}{
-					map[string]interface{}{
-						"type": "image_url",
-						"image_url": map[string]interface{}{
-							"url": "data:image/jpeg;base64,old",
-						},
-					},
-				},
-			},
-			expected: "[orange::i][image: image][-:-:-]",
-		},
-		{
-			name:  "old format with path",
-			index: 5,
-			msg: RoleMsg{
-				Role:            "user",
-				hasContentParts: true,
-				ContentParts: []interface{}{
-					map[string]interface{}{
-						"type": "image_url",
-						"path": "/old/path/photo.jpg",
-						"image_url": map[string]interface{}{
-							"url": "data:image/jpeg;base64,old",
-						},
-					},
-				},
-			},
-			expected: "[orange::i][image: /old/path/photo.jpg][-:-:-]",
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			result := tt.msg.ToText(tt.index)
-			if !strings.Contains(result, tt.expected) {
-				t.Errorf("ToText() result does not contain expected indicator\ngot: %s\nwant substring: %s", result, tt.expected)
-			}
-			// Ensure the indicator appears before text content
-			if strings.Contains(tt.expected, "cat.jpg") && strings.Contains(result, "Look at this picture") {
-				indicatorPos := strings.Index(result, "[orange::i][image: /home/user/Pictures/cat.jpg][-:-:-]")
-				textPos := strings.Index(result, "Look at this picture")
-				if indicatorPos == -1 || textPos == -1 || indicatorPos >= textPos {
-					t.Errorf("image indicator should appear before text")
-				}
-			}
-		})
-	}
-}
-func TestExtractDisplayPath(t *testing.T) {
-	// Save original base dir
-	originalBaseDir := imageBaseDir
-	defer func() { imageBaseDir = originalBaseDir }()
-	tests := []struct {
-		name     string
-		baseDir  string
-		path     string
-		expected string
-	}{
-		{
-			name:     "no base dir shows full path",
-			baseDir:  "",
-			path:     "/home/user/images/cat.jpg",
-			expected: "/home/user/images/cat.jpg",
-		},
-		{
-			name:     "relative path within base dir",
-			baseDir:  "/home/user",
-			path:     "/home/user/images/cat.jpg",
-			expected: "images/cat.jpg",
-		},
-		{
-			name:     "path outside base dir shows full path",
-			baseDir:  "/home/user",
-			path:     "/tmp/test.jpg",
-			expected: "/tmp/test.jpg",
-		},
-		{
-			name:     "same directory",
-			baseDir:  "/home/user/images",
-			path:     "/home/user/images/cat.jpg",
-			expected: "cat.jpg",
-		},
-		{
-			name:     "long path truncated",
-			baseDir:  "",
-			path:     "/very/long/path/to/a/really_long_filename_that_exceeds_sixty_characters_limit_yes_it_is_very_long.jpg",
-			expected: "..._that_exceeds_sixty_characters_limit_yes_it_is_very_long.jpg",
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			imageBaseDir = tt.baseDir
-			result := extractDisplayPath(tt.path)
-			if result != tt.expected {
-				t.Errorf("extractDisplayPath(%q) with baseDir=%q = %q, want %q",
-					tt.path, tt.baseDir, result, tt.expected)
-			}
-		})
-	}
-}

models/openrouter.go

@@ -172,3 +172,16 @@ func (orm *ORModels) ListModels(free bool) []string {
 	}
 	return resp
 }
+
+func (orm *ORModels) HasVision(modelID string) bool {
+	for i := range orm.Data {
+		if orm.Data[i].ID == modelID {
+			for _, mod := range orm.Data[i].Architecture.InputModalities {
+				if mod == "image" {
+					return true
+				}
+			}
+		}
+	}
+	return false
+}

popups.go

@@ -1,6 +1,7 @@
 package main
 
 import (
+	"gf-lt/models"
 	"slices"
 	"strings"
 
@@ -51,7 +52,7 @@ func showModelSelectionPopup() {
 	// Find the current model index to set as selected
 	currentModelIndex := -1
 	for i, model := range modelList {
-		if strings.TrimPrefix(model, "(loaded) ") == chatBody.Model {
+		if strings.TrimPrefix(model, models.LoadedMark) == chatBody.Model {
 			currentModelIndex = i
 		}
 		modelListWidget.AddItem(model, "", 0, nil)
@@ -61,7 +62,7 @@ func showModelSelectionPopup() {
 		modelListWidget.SetCurrentItem(currentModelIndex)
 	}
 	modelListWidget.SetSelectedFunc(func(index int, mainText string, secondaryText string, shortcut rune) {
-		modelName := strings.TrimPrefix(mainText, "(loaded) ")
+		modelName := strings.TrimPrefix(mainText, models.LoadedMark)
 		chatBody.Model = modelName
 		cfg.CurrentModel = chatBody.Model
 		pages.RemovePage("modelSelectionPopup")
@@ -142,6 +143,7 @@ func showAPILinkSelectionPopup() {
 	apiListWidget.SetSelectedFunc(func(index int, mainText string, secondaryText string, shortcut rune) {
 		// Update the API in config
 		cfg.CurrentAPI = mainText
+		UpdateToolCapabilities()
 		// Update model list based on new API
 		// Helper function to get model list for a given API (same as in props_table.go)
 		getModelListForAPI := func(api string) []string {
@@ -159,8 +161,9 @@ func showAPILinkSelectionPopup() {
 		newModelList := getModelListForAPI(cfg.CurrentAPI)
 		// Ensure chatBody.Model is in the new list; if not, set to first available model
 		if len(newModelList) > 0 && !slices.Contains(newModelList, chatBody.Model) {
-			chatBody.Model = newModelList[0]
+			chatBody.Model = strings.TrimPrefix(newModelList[0], models.LoadedMark)
 			cfg.CurrentModel = chatBody.Model
+			UpdateToolCapabilities()
 		}
 		pages.RemovePage("apiLinkSelectionPopup")
 		app.SetFocus(textArea)

session.go

@@ -131,13 +131,18 @@ func loadOldChatOrGetNew() []models.RoleMsg {
 	chat, err := store.GetLastChat()
 	if err != nil {
 		logger.Warn("failed to load history chat", "error", err)
+		maxID, err := store.ChatGetMaxID()
+		if err != nil {
+			logger.Error("failed to fetch max chat id", "error", err)
+		}
+		maxID++
 		chat := &models.Chat{
-			ID:        0,
+			ID:        maxID,
 			CreatedAt: time.Now(),
 			UpdatedAt: time.Now(),
 			Agent:     cfg.AssistantRole,
 		}
-		chat.Name = fmt.Sprintf("%s_%v", chat.Agent, chat.CreatedAt.Unix())
+		chat.Name = fmt.Sprintf("%s_%v", chat.Agent, chat.ID)
 		activeChatName = chat.Name
 		chatMap[chat.Name] = chat
 		return defaultStarter
@@ -149,10 +154,6 @@ func loadOldChatOrGetNew() []models.RoleMsg {
 		chatMap[chat.Name] = chat
 		return defaultStarter
 	}
-	// if chat.Name == "" {
-	// 	logger.Warn("empty chat name", "id", chat.ID)
-	// 	chat.Name = fmt.Sprintf("%s_%v", chat.Agent, chat.CreatedAt.Unix())
-	// }
 	chatMap[chat.Name] = chat
 	activeChatName = chat.Name
 	cfg.AssistantRole = chat.Agent

sysprompts/coding_assistant.json

@@ -1,6 +1,6 @@
 {
-  "sys_prompt": "You are an expert software engineering assistant. Your goal is to help users with coding tasks, debugging, refactoring, and software development.\n\n## Core Principles\n1. **Security First**: Never expose secrets, keys, or credentials. Never commit sensitive data.\n2. **No Git Actions**: You can READ git info (status, log, diff) for context, but NEVER perform git actions (commit, add, push, checkout, reset, rm, etc.). Let the user handle all git operations.\n3. **Explore Before Execute**: Always understand the codebase structure before making changes.\n4. **Follow Conventions**: Match existing code style, patterns, and frameworks used in the project.\n5. **Be Concise**: Minimize output tokens while maintaining quality. Avoid unnecessary explanations.\n\n## Workflow for Complex Tasks\nFor multi-step tasks, ALWAYS use the todo system to track progress:\n\n1. **Create Todo List**: At the start of complex tasks, use `todo_create` to break down work into actionable items.\n2. **Update Progress**: Mark items as `in_progress` when working on them, and `completed` when done.\n3. **Check Status**: Use `todo_read` to review your progress.\n\nExample workflow:\n- User: \"Add user authentication to this app\"\n- You: Create todos: [\"Analyze existing auth structure\", \"Check frameworks in use\", \"Implement auth middleware\", \"Add login endpoints\", \"Test implementation\"]\n\n## Task Execution Flow\n\n### Phase 1: Exploration (Always First)\n- Use `file_list` to understand directory structure (path defaults to FilePickerDir if not specified)\n- Use `file_read` to examine relevant files (paths are relative to FilePickerDir unless starting with `/`)\n- Use `execute_command` with `grep`/`find` to search for patterns\n- Check `README` or documentation files\n- Identify: frameworks, conventions, testing approach\n- **Git reads allowed**: You may use `git status`, `git log`, `git diff` for context, but only to inform your work\n- **Path handling**: Relative paths are resolved against FilePickerDir (configurable via Alt+O). Use absolute paths (starting with `/`) to bypass FilePickerDir.\n\n### Phase 2: Planning\n- For complex tasks: create todo items\n- Identify files that need modification\n- Plan your approach following existing patterns\n\n### Phase 3: Implementation\n- Make changes using appropriate file tools\n- Prefer `file_write` for new files, `file_read` then modify for existing files\n- Follow existing code style exactly\n- Use existing libraries and utilities\n\n### Phase 4: Verification\n- Run tests if available (check for test scripts)\n- Run linting/type checking commands\n- Verify changes work as expected\n\n### Phase 5: Completion\n- Update todos to `completed`\n- Provide concise summary of changes\n- Reference specific file paths and line numbers when relevant\n- **DO NOT commit changes** - inform user what was done so they can review and commit themselves\n\n## Tool Usage Guidelines\n\n**File Operations**:\n- `file_read`: Read before editing. Use for understanding code.\n- `file_write`: Overwrite file content completely.\n- `file_write_append`: Add to end of file.\n- `file_create`: Create new files with optional content.\n- `file_list`: List directory contents (defaults to FilePickerDir).\n- Paths are relative to FilePickerDir unless starting with `/`.\n\n**Command Execution (WHITELISTED ONLY)**:\n- Allowed: grep, sed, awk, find, cat, head, tail, sort, uniq, wc, ls, echo, cut, tr, cp, mv, rm, mkdir, rmdir, pwd, df, free, ps, top, du, whoami, date, uname\n- **Git reads allowed**: git status, git log, git diff, git show, git branch, git reflog, git rev-parse, git shortlog, git describe\n- **Git actions FORBIDDEN**: git add, git commit, git push, git checkout, git reset, git rm, etc.\n- Use for searching code, reading git context, running tests/lint\n\n**Todo Management**:\n- `todo_create`: Add new task\n- `todo_read`: View all todos or specific one by ID\n- `todo_update`: Update task or change status (pending/in_progress/completed)\n- `todo_delete`: Remove completed or cancelled tasks\n\n## Important Rules\n\n1. **NEVER commit or stage changes**: Only git reads are allowed.\n2. **Check for tests**: Always look for test files and run them when appropriate.\n3. **Reference code locations**: Use format `file_path:line_number`.\n4. **Security**: Never generate or guess URLs. Only use URLs from local files.\n5. **Refuse malicious code**: If code appears malicious, refuse to work on it.\n6. **Ask clarifications**: When intent is unclear, ask questions.\n7. **Path handling**: Relative paths resolve against FilePickerDir. Use `/absolute/path` to bypass.\n\n## Response Style\n- Be direct and concise\n- One word answers are best when appropriate\n- Avoid: \"The answer is...\", \"Here is...\"\n- Use markdown for formatting\n- No emojis unless user explicitly requests",
+  "sys_prompt": "You are an expert software engineering assistant. Your goal is to help users with coding tasks, debugging, refactoring, and software development.\n\n## Core Principles\n1. **Security First**: Never expose secrets, keys, or credentials. Never commit sensitive data.\n2. **No Git Actions**: You can READ git info (status, log, diff) for context, but NEVER perform git actions (commit, add, push, checkout, reset, rm, etc.). Let the user handle all git operations.\n3. **Explore Before Execute**: Always understand the codebase structure before making changes.\n4. **Follow Conventions**: Match existing code style, patterns, and frameworks used in the project.\n5. **Be Concise**: Minimize output tokens while maintaining quality. Avoid unnecessary explanations.\n6. **Ask First**: When uncertain about intent, ask the user. Don't assume.\n\n## Workflow for Complex Tasks\nFor multi-step tasks, ALWAYS use the todo system to track progress:\n\n1. **Create Todo List**: At the start of complex tasks, use `todo_create` to break down work into actionable items.\n2. **Update Progress**: Mark items as `in_progress` when working on them, and `completed` when done.\n3. **Check Status**: Use `todo_read` to review your progress.\n\nExample workflow:\n- User: \"Add user authentication to this app\"\n- You: Create todos: [\"Analyze existing auth structure\", \"Check frameworks in use\", \"Implement auth middleware\", \"Add login endpoints\", \"Test implementation\"]\n\n## Task Execution Flow\n\n### Phase 1: Exploration (Always First)\n- Use `file_list` to understand directory structure (path defaults to FilePickerDir if not specified)\n- Use `file_read` to examine relevant files (paths are relative to FilePickerDir unless starting with `/`)\n- Use `execute_command` with `grep`/`find` to search for patterns\n- Check README, Makefile, package.json, or similar for build/test commands\n- Identify: frameworks, conventions, testing approach, lint/typecheck commands\n- **Git reads allowed**: You may use `git status`, `git log`, `git diff` for context, but only to inform your work\n- **Path handling**: Relative paths resolve against FilePickerDir; absolute paths (starting with `/`) bypass it\n\n### Phase 2: Planning\n- For complex tasks: create todo items\n- Identify files that need modification\n- Plan your approach following existing patterns\n\n### Phase 3: Implementation\n- Make changes using appropriate file tools\n- Prefer `file_write` for new files, `file_read` then edit for existing files\n- Follow existing code style exactly\n- Use existing libraries and utilities\n\n### Phase 4: Verification\n- Run tests if available (check for test scripts in README/Makefile)\n- Run linting/type checking commands\n- Verify changes work as expected\n\n### Phase 5: Completion\n- Update todos to `completed`\n- Provide concise summary of changes\n- Reference specific file paths and line numbers when relevant\n- **DO NOT commit changes** - inform user what was done so they can review and commit themselves\n\n## Command Execution\n- Use `execute_command` with a single string containing command and arguments (e.g., `go run main.go`, `ls -la`, `cd /tmp`)\n- Use `cd /path` to change the working directory for file operations",
   "role": "CodingAssistant",
   "filepath": "sysprompts/coding_assistant.json",
-  "first_msg": "Hello! I'm your coding assistant. I can help you with software engineering tasks like writing code, debugging, refactoring, and exploring codebases. I work best when you give me specific tasks, and for complex work, I'll create a todo list to track my progress. What would you like to work on?"
+  "first_msg": "Hello! I'm your coding assistant. Give me a specific task and I'll get started. For complex work, I'll track progress with todos."
 }

tools.go

@@ -85,6 +85,11 @@ Your current tools:
 "when_to_use": "when asked to read the content of a file"
 },
 {
+"name":"file_read_image",
+"args": ["path"],
+"when_to_use": "when asked to read or view an image file"
+},
+{
 "name":"file_write",
 "args": ["path", "content"],
 "when_to_use": "when needed to overwrite content to a file"
@@ -95,6 +100,11 @@ Your current tools:
 "when_to_use": "when asked to append content to a file; use sed to edit content"
 },
 {
+"name":"file_edit",
+"args": ["path", "oldString", "newString", "lineNumber"],
+"when_to_use": "when you need to make targeted changes to a specific section of a file without rewriting the entire file; lineNumber is optional - if provided, only edits that specific line; if not provided, replaces all occurrences of oldString"
+},
+{
 "name":"file_delete",
 "args": ["path"],
 "when_to_use": "when asked to delete a file"
@@ -117,7 +127,7 @@ Your current tools:
 {
 "name":"execute_command",
 "args": ["command", "args"],
-"when_to_use": "when asked to execute a system command; args is optional; allowed commands: grep, sed, awk, find, cat, head, tail, sort, uniq, wc, ls, echo, cut, tr, cp, mv, rm, mkdir, rmdir, pwd, df, free, ps, top, du, whoami, date, uname"
+"when_to_use": "when asked to execute a system command; args is optional; allowed commands: grep, sed, awk, find, cat, head, tail, sort, uniq, wc, ls, echo, cut, tr, cp, mv, rm, mkdir, rmdir, pwd, df, free, ps, top, du, whoami, date, uname, go"
 }
 ]
 </tools>
@@ -165,8 +175,36 @@ After that you are free to respond to the user.
 	webAgentsOnce      sync.Once
 )
 
+var windowToolSysMsg = `
+Additional window tools (available only if xdotool and maim are installed):
+[
+{
+"name":"list_windows",
+"args": [],
+"when_to_use": "when asked to list visible windows; returns map of window ID to window name"
+},
+{
+"name":"capture_window",
+"args": ["window"],
+"when_to_use": "when asked to take a screenshot of a specific window; saves to /tmp; window can be ID or name substring; returns file path"
+},
+{
+"name":"capture_window_and_view",
+"args": ["window"],
+"when_to_use": "when asked to take a screenshot of a specific window and show it; saves to /tmp and returns image for viewing; window can be ID or name substring"
+}
+]
+`
+
 var WebSearcher searcher.WebSurfer
 
+var (
+	windowToolsAvailable bool
+	xdotoolPath          string
+	maimPath             string
+	modelHasVision       bool
+)
+
 func init() {
 	sa, err := searcher.NewWebSurfer(searcher.SearcherTypeScraper, "")
 	if err != nil {
@@ -176,6 +214,47 @@ func init() {
 	if err := rag.Init(cfg, logger, store); err != nil {
 		logger.Warn("failed to init rag; rag_search tool will not be available", "error", err)
 	}
+	checkWindowTools()
+	registerWindowTools()
+}
+
+func checkWindowTools() {
+	xdotoolPath, _ = exec.LookPath("xdotool")
+	maimPath, _ = exec.LookPath("maim")
+	windowToolsAvailable = xdotoolPath != "" && maimPath != ""
+	if windowToolsAvailable {
+		logger.Info("window tools available: xdotool and maim found")
+	} else {
+		if xdotoolPath == "" {
+			logger.Warn("xdotool not found, window listing tools will not be available")
+		}
+		if maimPath == "" {
+			logger.Warn("maim not found, window capture tools will not be available")
+		}
+	}
+}
+
+func UpdateToolCapabilities() {
+	if !cfg.ToolUse {
+		return
+	}
+	modelHasVision = false
+	if cfg == nil || cfg.CurrentAPI == "" {
+		logger.Warn("cannot determine model capabilities: cfg or CurrentAPI is nil")
+		registerWindowTools()
+		return
+	}
+	prevHasVision := modelHasVision
+	modelHasVision = ModelHasVision(cfg.CurrentAPI, cfg.CurrentModel)
+	if modelHasVision {
+		logger.Info("model has vision support", "model", cfg.CurrentModel, "api", cfg.CurrentAPI)
+	} else {
+		logger.Info("model does not have vision support", "model", cfg.CurrentModel, "api", cfg.CurrentAPI)
+		if windowToolsAvailable && !prevHasVision && !modelHasVision {
+			_ = notifyUser("window tools", "Window capture-and-view unavailable: model lacks vision support")
+		}
+	}
+	registerWindowTools()
 }
 
 // getWebAgentClient returns a singleton AgentClient for web agents.
@@ -464,6 +543,43 @@ func fileRead(args map[string]string) []byte {
 	return jsonResult
 }
 
+func fileReadImage(args map[string]string) []byte {
+	path, ok := args["path"]
+	if !ok || path == "" {
+		msg := "path not provided to file_read_image tool"
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	path = resolvePath(path)
+	dataURL, err := models.CreateImageURLFromPath(path)
+	if err != nil {
+		msg := "failed to read image; error: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	// result := map[string]any{
+	// 	"type": "multimodal_content",
+	// 	"parts": []map[string]string{
+	// 		{"type": "text", "text": "Image at " + path},
+	// 		{"type": "image_url", "url": dataURL},
+	// 	},
+	// }
+	result := models.MultimodalToolResp{
+		Type: "multimodal_content",
+		Parts: []map[string]string{
+			{"type": "text", "text": "Image at " + path},
+			{"type": "image_url", "url": dataURL},
+		},
+	}
+	jsonResult, err := json.Marshal(result)
+	if err != nil {
+		msg := "failed to marshal result; error: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	return jsonResult
+}
+
 func fileWrite(args map[string]string) []byte {
 	path, ok := args["path"]
 	if !ok || path == "" {
@@ -506,6 +622,77 @@ func fileWriteAppend(args map[string]string) []byte {
 	return []byte(msg)
 }
 
+func fileEdit(args map[string]string) []byte {
+	path, ok := args["path"]
+	if !ok || path == "" {
+		msg := "path not provided to file_edit tool"
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	path = resolvePath(path)
+	oldString, ok := args["oldString"]
+	if !ok || oldString == "" {
+		msg := "oldString not provided to file_edit tool"
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	newString, ok := args["newString"]
+	if !ok {
+		newString = ""
+	}
+	lineNumberStr, hasLineNumber := args["lineNumber"]
+	// Read file content
+	content, err := os.ReadFile(path)
+	if err != nil {
+		msg := "failed to read file: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	fileContent := string(content)
+	var replacementCount int
+	if hasLineNumber && lineNumberStr != "" {
+		// Line-number based edit
+		lineNum, err := strconv.Atoi(lineNumberStr)
+		if err != nil {
+			msg := "invalid lineNumber: must be a valid integer"
+			logger.Error(msg)
+			return []byte(msg)
+		}
+		lines := strings.Split(fileContent, "\n")
+		if lineNum < 1 || lineNum > len(lines) {
+			msg := fmt.Sprintf("lineNumber %d out of range (file has %d lines)", lineNum, len(lines))
+			logger.Error(msg)
+			return []byte(msg)
+		}
+		// Find oldString in the specific line
+		targetLine := lines[lineNum-1]
+		if !strings.Contains(targetLine, oldString) {
+			msg := fmt.Sprintf("oldString not found on line %d", lineNum)
+			logger.Error(msg)
+			return []byte(msg)
+		}
+		lines[lineNum-1] = strings.Replace(targetLine, oldString, newString, 1)
+		replacementCount = 1
+		fileContent = strings.Join(lines, "\n")
+	} else {
+		// Replace all occurrences
+		if !strings.Contains(fileContent, oldString) {
+			msg := "oldString not found in file"
+			logger.Error(msg)
+			return []byte(msg)
+		}
+		fileContent = strings.ReplaceAll(fileContent, oldString, newString)
+		replacementCount = strings.Count(fileContent, newString)
+	}
+	if err := os.WriteFile(path, []byte(fileContent), 0644); err != nil {
+		msg := "failed to write file: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	msg := fmt.Sprintf("file edited successfully at %s (%d replacement(s))", path, replacementCount)
+	return []byte(msg)
+}
+
 func fileDelete(args map[string]string) []byte {
 	path, ok := args["path"]
 	if !ok || path == "" {
@@ -681,35 +868,31 @@ func listDirectory(path string) ([]string, error) {
 
 // Command Execution Tool
 func executeCommand(args map[string]string) []byte {
-	command, ok := args["command"]
+	commandStr := args["command"]
-	if !ok || command == "" {
+	if commandStr == "" {
 		msg := "command not provided to execute_command tool"
 		logger.Error(msg)
 		return []byte(msg)
 	}
-	// Get arguments - handle both single arg and multiple args
+	// Handle commands passed as single string with spaces (e.g., "go run main.go" or "cd /tmp")
-	var cmdArgs []string
+	// Split into base command and arguments
-	if args["args"] != "" {
+	parts := strings.Fields(commandStr)
-		// If args is provided as a single string, split by spaces
+	if len(parts) == 0 {
-		cmdArgs = strings.Fields(args["args"])
+		msg := "command not provided to execute_command tool"
-	} else {
+		logger.Error(msg)
-		// If individual args are provided, collect them
+		return []byte(msg)
-		argNum := 1
-		for {
-			argKey := fmt.Sprintf("arg%d", argNum)
-			if argValue, exists := args[argKey]; exists && argValue != "" {
-				cmdArgs = append(cmdArgs, argValue)
-			} else {
-				break
-			}
-			argNum++
-		}
 	}
+	command := parts[0]
+	cmdArgs := parts[1:]
 	if !isCommandAllowed(command, cmdArgs...) {
 		msg := fmt.Sprintf("command '%s' is not allowed", command)
 		logger.Error(msg)
 		return []byte(msg)
 	}
+	// Special handling for cd command - update FilePickerDir
+	if command == "cd" {
+		return handleCdCommand(cmdArgs)
+	}
 	// Execute with timeout for safety
 	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 	defer cancel()
@@ -723,12 +906,58 @@ func executeCommand(args map[string]string) []byte {
 	}
 	// Check if output is empty and return success message
 	if len(output) == 0 {
-		successMsg := fmt.Sprintf("command '%s %s' executed successfully and exited with code 0", command, strings.Join(cmdArgs, " "))
+		successMsg := fmt.Sprintf("command '%s' executed successfully and exited with code 0", commandStr)
 		return []byte(successMsg)
 	}
 	return output
 }
 
+// handleCdCommand handles the cd command to update FilePickerDir
+func handleCdCommand(args []string) []byte {
+	var targetDir string
+	if len(args) == 0 {
+		// cd with no args goes to home directory
+		homeDir, err := os.UserHomeDir()
+		if err != nil {
+			msg := "cd: cannot determine home directory: " + err.Error()
+			logger.Error(msg)
+			return []byte(msg)
+		}
+		targetDir = homeDir
+	} else {
+		targetDir = args[0]
+	}
+
+	// Resolve relative paths against current FilePickerDir
+	if !filepath.IsAbs(targetDir) {
+		targetDir = filepath.Join(cfg.FilePickerDir, targetDir)
+	}
+
+	// Verify the directory exists
+	info, err := os.Stat(targetDir)
+	if err != nil {
+		msg := "cd: " + targetDir + ": " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	if !info.IsDir() {
+		msg := "cd: " + targetDir + ": not a directory"
+		logger.Error(msg)
+		return []byte(msg)
+	}
+
+	// Update FilePickerDir
+	absDir, err := filepath.Abs(targetDir)
+	if err != nil {
+		msg := "cd: failed to resolve path: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	cfg.FilePickerDir = absDir
+	msg := "FilePickerDir changed to: " + absDir
+	return []byte(msg)
+}
+
 // Helper functions for command execution
 // Todo structure
 type TodoItem struct {
@@ -740,6 +969,14 @@ type TodoList struct {
 	Items []TodoItem `json:"items"`
 }
 
+func (t TodoList) ToString() string {
+	sb := strings.Builder{}
+	for i := range t.Items {
+		fmt.Fprintf(&sb, "\n[%s] %s. %s\n", t.Items[i].Status, t.Items[i].ID, t.Items[i].Task)
+	}
+	return sb.String()
+}
+
 // Global todo list storage
 var globalTodoList = TodoList{
 	Items: []TodoItem{},
@@ -766,6 +1003,7 @@ func todoCreate(args map[string]string) []byte {
 		"id":      id,
 		"task":    task,
 		"status":  "pending",
+		"todos":   globalTodoList.ToString(),
 	}
 	jsonResult, err := json.Marshal(result)
 	if err != nil {
@@ -777,38 +1015,9 @@ func todoCreate(args map[string]string) []byte {
 }
 
 func todoRead(args map[string]string) []byte {
-	id, ok := args["id"]
-	if ok && id != "" {
-		// Find specific todo by ID
-		for _, item := range globalTodoList.Items {
-			if item.ID == id {
-				result := map[string]interface{}{
-					"todo": item,
-				}
-				jsonResult, err := json.Marshal(result)
-				if err != nil {
-					msg := "failed to marshal result; error: " + err.Error()
-					logger.Error(msg)
-					return []byte(msg)
-				}
-				return jsonResult
-			}
-		}
-		// ID not found
-		result := map[string]string{
-			"error": fmt.Sprintf("todo with id %s not found", id),
-		}
-		jsonResult, err := json.Marshal(result)
-		if err != nil {
-			msg := "failed to marshal result; error: " + err.Error()
-			logger.Error(msg)
-			return []byte(msg)
-		}
-		return jsonResult
-	}
 	// Return all todos if no ID specified
 	result := map[string]interface{}{
-		"todos": globalTodoList.Items,
+		"todos": globalTodoList.ToString(),
 	}
 	jsonResult, err := json.Marshal(result)
 	if err != nil {
@@ -859,6 +1068,7 @@ func todoUpdate(args map[string]string) []byte {
 			result := map[string]string{
 				"message": "todo updated successfully",
 				"id":      id,
+				"todos":   globalTodoList.ToString(),
 			}
 			jsonResult, err := json.Marshal(result)
 			if err != nil {
@@ -897,6 +1107,7 @@ func todoDelete(args map[string]string) []byte {
 			result := map[string]string{
 				"message": "todo deleted successfully",
 				"id":      id,
+				"todos":   globalTodoList.ToString(),
 			}
 			jsonResult, err := json.Marshal(result)
 			if err != nil {
@@ -934,6 +1145,7 @@ var gitReadSubcommands = map[string]bool{
 
 func isCommandAllowed(command string, args ...string) bool {
 	allowedCommands := map[string]bool{
+		"cd":     true,
 		"grep":   true,
 		"sed":    true,
 		"awk":    true,
@@ -963,13 +1175,18 @@ func isCommandAllowed(command string, args ...string) bool {
 		"date":   true,
 		"uname":  true,
 		"git":    true,
+		"go":     true,
 	}
-	if !allowedCommands[command] {
+	// Allow all go subcommands (go run, go mod tidy, go test, etc.)
-		return false
+	if strings.HasPrefix(command, "go ") && allowedCommands["go"] {
+		return true
 	}
 	if command == "git" && len(args) > 0 {
 		return gitReadSubcommands[args[0]]
 	}
+	if !allowedCommands[command] {
+		return false
+	}
 	return true
 }
 
@@ -982,6 +1199,142 @@ func summarizeChat(args map[string]string) []byte {
 	return []byte(chatText)
 }
 
+func windowIDToHex(decimalID string) string {
+	id, err := strconv.ParseInt(decimalID, 10, 64)
+	if err != nil {
+		return decimalID
+	}
+	return fmt.Sprintf("0x%x", id)
+}
+
+func listWindows(args map[string]string) []byte {
+	if !windowToolsAvailable {
+		return []byte("window tools not available: xdotool or maim not found")
+	}
+	cmd := exec.Command(xdotoolPath, "search", "--name", ".")
+	output, err := cmd.Output()
+	if err != nil {
+		msg := "failed to list windows: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	windowIDs := strings.Fields(string(output))
+	windows := make(map[string]string)
+	for _, id := range windowIDs {
+		id = strings.TrimSpace(id)
+		if id == "" {
+			continue
+		}
+		nameCmd := exec.Command(xdotoolPath, "getwindowname", id)
+		nameOutput, err := nameCmd.Output()
+		if err != nil {
+			continue
+		}
+		name := strings.TrimSpace(string(nameOutput))
+		windows[id] = name
+	}
+	data, err := json.Marshal(windows)
+	if err != nil {
+		msg := "failed to marshal window list: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	return data
+}
+
+func captureWindow(args map[string]string) []byte {
+	if !windowToolsAvailable {
+		return []byte("window tools not available: xdotool or maim not found")
+	}
+	window, ok := args["window"]
+	if !ok || window == "" {
+		return []byte("window parameter required (window ID or name)")
+	}
+	var windowID string
+	if _, err := strconv.Atoi(window); err == nil {
+		windowID = window
+	} else {
+		cmd := exec.Command(xdotoolPath, "search", "--name", window)
+		output, err := cmd.Output()
+		if err != nil || len(strings.Fields(string(output))) == 0 {
+			return []byte("window not found: " + window)
+		}
+		windowID = strings.Fields(string(output))[0]
+	}
+	nameCmd := exec.Command(xdotoolPath, "getwindowname", windowID)
+	nameOutput, _ := nameCmd.Output()
+	windowName := strings.TrimSpace(string(nameOutput))
+	windowName = regexp.MustCompile(`[^a-zA-Z]+`).ReplaceAllString(windowName, "")
+	if windowName == "" {
+		windowName = "window"
+	}
+	timestamp := time.Now().Unix()
+	filename := fmt.Sprintf("/tmp/%s_%d.jpg", windowName, timestamp)
+	cmd := exec.Command(maimPath, "-i", windowIDToHex(windowID), filename)
+	if err := cmd.Run(); err != nil {
+		msg := "failed to capture window: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	return []byte("screenshot saved: " + filename)
+}
+
+func captureWindowAndView(args map[string]string) []byte {
+	if !windowToolsAvailable {
+		return []byte("window tools not available: xdotool or maim not found")
+	}
+	window, ok := args["window"]
+	if !ok || window == "" {
+		return []byte("window parameter required (window ID or name)")
+	}
+	var windowID string
+	if _, err := strconv.Atoi(window); err == nil {
+		windowID = window
+	} else {
+		cmd := exec.Command(xdotoolPath, "search", "--name", window)
+		output, err := cmd.Output()
+		if err != nil || len(strings.Fields(string(output))) == 0 {
+			return []byte("window not found: " + window)
+		}
+		windowID = strings.Fields(string(output))[0]
+	}
+	nameCmd := exec.Command(xdotoolPath, "getwindowname", windowID)
+	nameOutput, _ := nameCmd.Output()
+	windowName := strings.TrimSpace(string(nameOutput))
+	windowName = regexp.MustCompile(`[^a-zA-Z]+`).ReplaceAllString(windowName, "")
+	if windowName == "" {
+		windowName = "window"
+	}
+	timestamp := time.Now().Unix()
+	filename := fmt.Sprintf("/tmp/%s_%d.jpg", windowName, timestamp)
+	captureCmd := exec.Command(maimPath, "-i", windowIDToHex(windowID), filename)
+	if err := captureCmd.Run(); err != nil {
+		msg := "failed to capture window: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	dataURL, err := models.CreateImageURLFromPath(filename)
+	if err != nil {
+		msg := "failed to create image URL: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	result := models.MultimodalToolResp{
+		Type: "multimodal_content",
+		Parts: []map[string]string{
+			{"type": "text", "text": "Screenshot saved: " + filename},
+			{"type": "image_url", "url": dataURL},
+		},
+	}
+	jsonResult, err := json.Marshal(result)
+	if err != nil {
+		msg := "failed to marshal result: " + err.Error()
+		logger.Error(msg)
+		return []byte(msg)
+	}
+	return jsonResult
+}
+
 type fnSig func(map[string]string) []byte
 
 var fnMap = map[string]fnSig{
@@ -995,8 +1348,10 @@ var fnMap = map[string]fnSig{
 	"read_url_raw":      readURLRaw,
 	"file_create":       fileCreate,
 	"file_read":         fileRead,
+	"file_read_image":   fileReadImage,
 	"file_write":        fileWrite,
 	"file_write_append": fileWriteAppend,
+	"file_edit":         fileEdit,
 	"file_delete":       fileDelete,
 	"file_move":         fileMove,
 	"file_copy":         fileCopy,
@@ -1009,6 +1364,66 @@ var fnMap = map[string]fnSig{
 	"summarize_chat":    summarizeChat,
 }
 
+func registerWindowTools() {
+	if windowToolsAvailable {
+		fnMap["list_windows"] = listWindows
+		fnMap["capture_window"] = captureWindow
+		windowTools := []models.Tool{
+			{
+				Type: "function",
+				Function: models.ToolFunc{
+					Name:        "list_windows",
+					Description: "List all visible windows with their IDs and names. Returns a map of window ID to window name.",
+					Parameters: models.ToolFuncParams{
+						Type:       "object",
+						Required:   []string{},
+						Properties: map[string]models.ToolArgProps{},
+					},
+				},
+			},
+			{
+				Type: "function",
+				Function: models.ToolFunc{
+					Name:        "capture_window",
+					Description: "Capture a screenshot of a specific window and save it to /tmp. Requires window parameter (window ID or name substring).",
+					Parameters: models.ToolFuncParams{
+						Type:     "object",
+						Required: []string{"window"},
+						Properties: map[string]models.ToolArgProps{
+							"window": models.ToolArgProps{
+								Type:        "string",
+								Description: "window ID or window name (partial match)",
+							},
+						},
+					},
+				},
+			},
+		}
+		if modelHasVision {
+			fnMap["capture_window_and_view"] = captureWindowAndView
+			windowTools = append(windowTools, models.Tool{
+				Type: "function",
+				Function: models.ToolFunc{
+					Name:        "capture_window_and_view",
+					Description: "Capture a screenshot of a specific window, save it to /tmp, and return the image for viewing. Requires window parameter (window ID or name substring).",
+					Parameters: models.ToolFuncParams{
+						Type:     "object",
+						Required: []string{"window"},
+						Properties: map[string]models.ToolArgProps{
+							"window": models.ToolArgProps{
+								Type:        "string",
+								Description: "window ID or window name (partial match)",
+							},
+						},
+					},
+				},
+			})
+		}
+		baseTools = append(baseTools, windowTools...)
+		toolSysMsg += windowToolSysMsg
+	}
+}
+
 // callToolWithAgent calls the tool and applies any registered agent.
 func callToolWithAgent(name string, args map[string]string) []byte {
 	registerWebAgents()
@@ -1220,6 +1635,24 @@ var baseTools = []models.Tool{
 			},
 		},
 	},
+	// file_read_image
+	models.Tool{
+		Type: "function",
+		Function: models.ToolFunc{
+			Name:        "file_read_image",
+			Description: "Read an image file and return it for multimodal LLM viewing. Supports png, jpg, jpeg, gif, webp formats. Use when you need the LLM to see and analyze an image.",
+			Parameters: models.ToolFuncParams{
+				Type:     "object",
+				Required: []string{"path"},
+				Properties: map[string]models.ToolArgProps{
+					"path": models.ToolArgProps{
+						Type:        "string",
+						Description: "path of the image file to read",
+					},
+				},
+			},
+		},
+	},
 	// file_write
 	models.Tool{
 		Type: "function",
@@ -1264,6 +1697,36 @@ var baseTools = []models.Tool{
 			},
 		},
 	},
+	// file_edit
+	models.Tool{
+		Type: "function",
+		Function: models.ToolFunc{
+			Name:        "file_edit",
+			Description: "Edit a specific section of a file by replacing oldString with newString. Use for targeted changes without rewriting the entire file.",
+			Parameters: models.ToolFuncParams{
+				Type:     "object",
+				Required: []string{"path", "oldString", "newString"},
+				Properties: map[string]models.ToolArgProps{
+					"path": models.ToolArgProps{
+						Type:        "string",
+						Description: "path of the file to edit",
+					},
+					"oldString": models.ToolArgProps{
+						Type:        "string",
+						Description: "the exact string to find and replace",
+					},
+					"newString": models.ToolArgProps{
+						Type:        "string",
+						Description: "the string to replace oldString with",
+					},
+					"lineNumber": models.ToolArgProps{
+						Type:        "string",
+						Description: "optional line number (1-indexed) to edit - if provided, only that line is edited",
+					},
+				},
+			},
+		},
+	},
 	// file_delete
 	models.Tool{
 		Type: "function",
@@ -1349,18 +1812,14 @@ var baseTools = []models.Tool{
 		Type: "function",
 		Function: models.ToolFunc{
 			Name:        "execute_command",
-			Description: "Execute a shell command safely. Use when you need to run system commands like grep sed awk find cat head tail sort uniq wc ls echo cut tr cp mv rm mkdir rmdir pwd df free ps top du whoami date uname. Git is allowed for read-only operations: status, log, diff, show, branch, reflog, rev-parse, shortlog, describe.",
+			Description: "Execute a shell command safely. Use when you need to run system commands like cd grep sed awk find cat head tail sort uniq wc ls echo cut tr cp mv rm mkdir rmdir pwd df free ps top du whoami date uname go git. Git is allowed for read-only operations: status, log, diff, show, branch, reflog, rev-parse, shortlog, describe. Use 'cd /path' to change working directory.",
 			Parameters: models.ToolFuncParams{
 				Type:     "object",
 				Required: []string{"command"},
 				Properties: map[string]models.ToolArgProps{
 					"command": models.ToolArgProps{
 						Type:        "string",
-						Description: "command to execute (only commands from whitelist are allowed: grep sed awk find cat head tail sort uniq wc ls echo cut tr cp mv rm mkdir rmdir pwd df free ps top du whoami date uname; git allowed for reads: status log diff show branch reflog rev-parse shortlog describe)",
+						Description: "command to execute with arguments (e.g., 'go run main.go', 'ls -la /tmp', 'cd /home/user'). Use a single string; arguments should be space-separated after the command.",
-					},
-					"args": models.ToolArgProps{
-						Type:        "string",
-						Description: "command arguments as a single string (e.g., '-la {path}')",
 					},
 				},
 			},
@@ -1422,7 +1881,7 @@ var baseTools = []models.Tool{
 					},
 					"status": models.ToolArgProps{
 						Type:        "string",
-						Description: "new status for the todo: pending, in_progress, or completed (optional)",
+						Description: "new status: pending, in_progress, or completed (optional)",
 					},
 				},
 			},
@@ -1447,3 +1906,56 @@ var baseTools = []models.Tool{
 		},
 	},
 }
+
+func init() {
+	if windowToolsAvailable {
+		baseTools = append(baseTools,
+			models.Tool{
+				Type: "function",
+				Function: models.ToolFunc{
+					Name:        "list_windows",
+					Description: "List all visible windows with their IDs and names. Returns a map of window ID to window name.",
+					Parameters: models.ToolFuncParams{
+						Type:       "object",
+						Required:   []string{},
+						Properties: map[string]models.ToolArgProps{},
+					},
+				},
+			},
+			models.Tool{
+				Type: "function",
+				Function: models.ToolFunc{
+					Name:        "capture_window",
+					Description: "Capture a screenshot of a specific window and save it to /tmp. Requires window parameter (window ID or name substring).",
+					Parameters: models.ToolFuncParams{
+						Type:     "object",
+						Required: []string{"window"},
+						Properties: map[string]models.ToolArgProps{
+							"window": models.ToolArgProps{
+								Type:        "string",
+								Description: "window ID or window name (partial match)",
+							},
+						},
+					},
+				},
+			},
+			models.Tool{
+				Type: "function",
+				Function: models.ToolFunc{
+					Name:        "capture_window_and_view",
+					Description: "Capture a screenshot of a specific window, save it to /tmp, and return the image for viewing. Requires window parameter (window ID or name substring).",
+					Parameters: models.ToolFuncParams{
+						Type:     "object",
+						Required: []string{"window"},
+						Properties: map[string]models.ToolArgProps{
+							"window": models.ToolArgProps{
+								Type:        "string",
+								Description: "window ID or window name (partial match)",
+							},
+						},
+					},
+				},
+			},
+		)
+	}
+}

tui.go

@@ -35,6 +35,8 @@ var (
 	renameWindow       *tview.InputField
 	roleEditWindow     *tview.InputField
 	shellInput         *tview.InputField
+	confirmModal       *tview.Modal
+	confirmPageName    = "confirm"
 	fullscreenMode     bool
 	positionVisible    bool = true
 	scrollToEndEnabled bool = true
@@ -80,7 +82,7 @@ var (
 [yellow]Ctrl+p[white]: props edit form (min-p, dry, etc.)
 [yellow]Ctrl+v[white]: show API link selection popup to choose current API
 [yellow]Ctrl+r[white]: start/stop recording from your microphone (needs stt server or whisper binary)
-[yellow]Ctrl+t[white]: remove thinking (<think>) and tool messages from context (delete from chat)
+[yellow]Ctrl+t[white]: (un)collapse tool messages
 [yellow]Ctrl+l[white]: show model selection popup to choose current model
 [yellow]Ctrl+k[white]: switch tool use (recommend tool use to llm after user msg)
 [yellow]Ctrl+a[white]: interrupt tts (needs tts server)
@@ -99,6 +101,7 @@ var (
 [yellow]Alt+8[white]: show char img or last picked img
 [yellow]Alt+9[white]: warm up (load) selected llama.cpp model
 [yellow]Alt+t[white]: toggle thinking blocks visibility (collapse/expand <think> blocks)
+[yellow]Ctrl+t[white]: toggle tool call/response visibility (collapse/expand tool calls and non-shell tool responses)
 [yellow]Alt+i[white]: show colorscheme selection popup
 
 === scrolling chat window (some keys similar to vim) ===
@@ -194,6 +197,39 @@ func init() {
 		}
 		return event
 	})
+	confirmModal = tview.NewModal().
+		SetText("You are trying to send an empty message.\nIt makes sense if the last message in the chat is from you.\nAre you sure?").
+		AddButtons([]string{"Yes", "No"}).
+		SetButtonBackgroundColor(tcell.ColorBlack).
+		SetButtonTextColor(tcell.ColorWhite).
+		SetDoneFunc(func(buttonIndex int, buttonLabel string) {
+			if buttonLabel == "Yes" {
+				persona := cfg.UserRole
+				if cfg.WriteNextMsgAs != "" {
+					persona = cfg.WriteNextMsgAs
+				}
+				chatRoundChan <- &models.ChatRoundReq{Role: persona, UserMsg: ""}
+			} // In both Yes and No, go back to the main page
+			pages.SwitchToPage("main") // or whatever your main page is named
+		})
+	confirmModal.SetInputCapture(func(event *tcell.EventKey) *tcell.EventKey {
+		if event.Key() == tcell.KeyRune {
+			switch event.Rune() {
+			case 'y', 'Y':
+				persona := cfg.UserRole
+				if cfg.WriteNextMsgAs != "" {
+					persona = cfg.WriteNextMsgAs
+				}
+				chatRoundChan <- &models.ChatRoundReq{Role: persona, UserMsg: ""}
+				pages.SwitchToPage("main")
+				return nil
+			case 'n', 'N', 'x', 'X':
+				pages.SwitchToPage("main")
+				return nil
+			}
+		}
+		return event
+	})
 	textArea = tview.NewTextArea().
 		SetPlaceholder("input is multiline; press <Enter> to start the next line;\npress <Esc> to send the message.")
 	textArea.SetBorder(true).SetTitle("input")
@@ -563,6 +599,20 @@ func init() {
 			}
 			return nil
 		}
+		// Handle Ctrl+T to toggle tool call/response visibility
+		if event.Key() == tcell.KeyCtrlT {
+			toolCollapsed = !toolCollapsed
+			textView.SetText(chatToText(chatBody.Messages, cfg.ShowSys))
+			colorText()
+			status := "expanded"
+			if toolCollapsed {
+				status = "collapsed"
+			}
+			if err := notifyUser("tools", "Tool calls/responses "+status); err != nil {
+				logger.Error("failed to send notification", "error", err)
+			}
+			return nil
+		}
 		if event.Key() == tcell.KeyRune && event.Rune() == 'i' && event.Modifiers()&tcell.ModAlt != 0 {
 			if isFullScreenPageActive() {
 				return event
@@ -781,14 +831,6 @@ func init() {
 			showModelSelectionPopup()
 			return nil
 		}
-		if event.Key() == tcell.KeyCtrlT {
-			// clear context
-			// remove tools and thinking
-			removeThinking(chatBody)
-			textView.SetText(chatToText(chatBody.Messages, cfg.ShowSys))
-			colorText()
-			return nil
-		}
 		if event.Key() == tcell.KeyCtrlV {
 			if isFullScreenPageActive() {
 				return event
@@ -990,42 +1032,43 @@ func init() {
 				return nil
 			}
 			msgText := textArea.GetText()
-			if msgText != "" {
+			nl := "\n\n" // keep empty lines between messages
-				nl := "\n\n" // keep empty lines between messages
+			prevText := textView.GetText(true)
-				prevText := textView.GetText(true)
+			persona := cfg.UserRole
-				persona := cfg.UserRole
+			// strings.LastIndex()
-				// strings.LastIndex()
+			// newline is not needed is prev msg ends with one
-				// newline is not needed is prev msg ends with one
+			if strings.HasSuffix(prevText, nl) {
-				if strings.HasSuffix(prevText, nl) {
+				nl = ""
-					nl = ""
+			} else if strings.HasSuffix(prevText, "\n") {
-				} else if strings.HasSuffix(prevText, "\n") {
+				nl = "\n" // only one newline, add another
-					nl = "\n" // only one newline, add another
-				}
-				if msgText != "" {
-					// as what char user sends msg?
-					if cfg.WriteNextMsgAs != "" {
-						persona = cfg.WriteNextMsgAs
-					}
-					// check if plain text
-					if !injectRole {
-						matches := roleRE.FindStringSubmatch(msgText)
-						if len(matches) > 1 {
-							persona = matches[1]
-							msgText = strings.TrimLeft(msgText[len(matches[0]):], " ")
-						}
-					}
-					// add user icon before user msg
-					fmt.Fprintf(textView, "%s[-:-:b](%d) <%s>: [-:-:-]\n%s\n",
-						nl, len(chatBody.Messages), persona, msgText)
-					textArea.SetText("", true)
-					if scrollToEndEnabled {
-						textView.ScrollToEnd()
-					}
-					colorText()
-				}
-				// go chatRound(msgText, persona, textView, false, false)
-				chatRoundChan <- &models.ChatRoundReq{Role: persona, UserMsg: msgText}
 			}
+			if msgText != "" {
+				// as what char user sends msg?
+				if cfg.WriteNextMsgAs != "" {
+					persona = cfg.WriteNextMsgAs
+				}
+				// check if plain text
+				if !injectRole {
+					matches := roleRE.FindStringSubmatch(msgText)
+					if len(matches) > 1 {
+						persona = matches[1]
+						msgText = strings.TrimLeft(msgText[len(matches[0]):], " ")
+					}
+				}
+				// add user icon before user msg
+				fmt.Fprintf(textView, "%s[-:-:b](%d) <%s>: [-:-:-]\n%s\n",
+					nl, len(chatBody.Messages), persona, msgText)
+				textArea.SetText("", true)
+				if scrollToEndEnabled {
+					textView.ScrollToEnd()
+				}
+				colorText()
+			} else {
+				pages.AddPage(confirmPageName, confirmModal, true, true)
+				return nil
+			}
+			// go chatRound(msgText, persona, textView, false, false)
+			chatRoundChan <- &models.ChatRoundReq{Role: persona, UserMsg: msgText}
 			return nil
 		}
 		if event.Key() == tcell.KeyPgUp || event.Key() == tcell.KeyPgDn {