9 Commits

Author SHA1 Message Date
Grail Finder
f40f09390b Feat(tts) alt+0 to replay last message in the chat 2026-02-18 13:15:40 +03:00
Grail Finder
5548991f5c Enha: statusline color (loaded:green; unloaded:red) for local models 2026-02-18 12:13:22 +03:00
Grail Finder
c12311da99 Chore: linter complaints 2026-02-18 08:42:05 +03:00
Grail Finder
7d18a9d77e Feat: indicator for a message with an image [image: filename] 2026-02-17 16:19:33 +03:00
Grail Finder
b67ae1be98 Enha: filter out thinking blocks from chat history, removed {role}: 2026-02-17 13:42:49 +03:00
Grail Finder
372e49199b Feat: collapse/expand thinking blocks with alt+t 2026-02-17 13:15:09 +03:00
Grail Finder
d6d4f09f8d Merge branch 'feat/filepicker-search' 2026-02-17 11:17:17 +03:00
Grail Finder
fa846225ee Enha: remove updatequeue, since it waits for another main action 2026-02-17 10:29:28 +03:00
Grail Finder
7b2fa04391 Fix (img prompt): botname: after <__media__> for /completion 2026-02-17 08:23:08 +03:00
13 changed files with 431 additions and 119 deletions

1
.gitignore vendored
View File

@@ -5,7 +5,6 @@ history/
*.db *.db
config.toml config.toml
sysprompts/* sysprompts/*
!sysprompts/cluedo.json
!sysprompts/alice_bob_carl.json !sysprompts/alice_bob_carl.json
history_bak/ history_bak/
.aider* .aider*

19
bot.go
View File

@@ -1087,7 +1087,15 @@ func chatToTextSlice(messages []models.RoleMsg, showSys bool) []string {
func chatToText(messages []models.RoleMsg, showSys bool) string { func chatToText(messages []models.RoleMsg, showSys bool) string {
s := chatToTextSlice(messages, showSys) s := chatToTextSlice(messages, showSys)
return strings.Join(s, "\n") text := strings.Join(s, "\n")
// Collapse thinking blocks if enabled
if thinkingCollapsed {
placeholder := "[yellow::i][thinking... (press Alt+T to expand)][-:-:-]"
text = thinkRE.ReplaceAllString(text, placeholder)
}
return text
} }
func removeThinking(chatBody *models.ChatBody) { func removeThinking(chatBody *models.ChatBody) {
@@ -1228,6 +1236,15 @@ func init() {
os.Exit(1) os.Exit(1)
return 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{ defaultStarter = []models.RoleMsg{
{Role: "system", Content: basicSysMsg}, {Role: "system", Content: basicSysMsg},
{Role: cfg.AssistantRole, Content: defaultFirstMsg}, {Role: cfg.AssistantRole, Content: defaultFirstMsg},

View File

@@ -48,3 +48,4 @@ EnableMouse = false # Enable mouse support in the UI
CharSpecificContextEnabled = true CharSpecificContextEnabled = true
CharSpecificContextTag = "@" CharSpecificContextTag = "@"
AutoTurn = true AutoTurn = true
StripThinkingFromAPI = true # Strip <think> blocks from messages before sending to LLM (keeps them in chat history)

View File

@@ -19,6 +19,7 @@ type Config struct {
ToolRole string `toml:"ToolRole"` ToolRole string `toml:"ToolRole"`
ToolUse bool `toml:"ToolUse"` ToolUse bool `toml:"ToolUse"`
ThinkUse bool `toml:"ThinkUse"` ThinkUse bool `toml:"ThinkUse"`
StripThinkingFromAPI bool `toml:"StripThinkingFromAPI"`
AssistantRole string `toml:"AssistantRole"` AssistantRole string `toml:"AssistantRole"`
SysDir string `toml:"SysDir"` SysDir string `toml:"SysDir"`
ChunkLimit uint32 `toml:"ChunkLimit"` ChunkLimit uint32 `toml:"ChunkLimit"`

View File

@@ -13,10 +13,9 @@ import (
"log/slog" "log/slog"
"net/http" "net/http"
"os" "os"
"regexp"
"strings" "strings"
"time"
"sync" "sync"
"time"
google_translate_tts "github.com/GrailFinder/google-translate-tts" google_translate_tts "github.com/GrailFinder/google-translate-tts"
"github.com/GrailFinder/google-translate-tts/handlers" "github.com/GrailFinder/google-translate-tts/handlers"
@@ -31,43 +30,8 @@ var (
TTSFlushChan = make(chan bool, 1) TTSFlushChan = make(chan bool, 1)
TTSDoneChan = make(chan bool, 1) TTSDoneChan = make(chan bool, 1)
// endsWithPunctuation = regexp.MustCompile(`[;.!?]$`) // endsWithPunctuation = regexp.MustCompile(`[;.!?]$`)
threeOrMoreDashesRE = regexp.MustCompile(`-{3,}`)
) )
// cleanText removes markdown and special characters that are not suitable for TTS
func cleanText(text string) string {
// Remove markdown-like characters that might interfere with TTS
text = strings.ReplaceAll(text, "*", "") // Bold/italic markers
text = strings.ReplaceAll(text, "#", "") // Headers
text = strings.ReplaceAll(text, "_", "") // Underline/italic markers
text = strings.ReplaceAll(text, "~", "") // Strikethrough markers
text = strings.ReplaceAll(text, "`", "") // Code markers
text = strings.ReplaceAll(text, "[", "") // Link brackets
text = strings.ReplaceAll(text, "]", "") // Link brackets
text = strings.ReplaceAll(text, "!", "") // Exclamation marks (if not punctuation)
// Remove HTML tags using regex
htmlTagRegex := regexp.MustCompile(`<[^>]*>`)
text = htmlTagRegex.ReplaceAllString(text, "")
// Split text into lines to handle table separators
lines := strings.Split(text, "\n")
var filteredLines []string
for _, line := range lines {
// Check if the line looks like a table separator (e.g., |----|, |===|, | - - - |)
// A table separator typically contains only |, -, =, and spaces
isTableSeparator := regexp.MustCompile(`^\s*\|\s*[-=\s]+\|\s*$`).MatchString(strings.TrimSpace(line))
if !isTableSeparator {
// If it's not a table separator, remove vertical bars but keep the content
processedLine := strings.ReplaceAll(line, "|", "")
filteredLines = append(filteredLines, processedLine)
}
// If it is a table separator, skip it (don't add to filteredLines)
}
text = strings.Join(filteredLines, "\n")
text = threeOrMoreDashesRE.ReplaceAllString(text, "")
text = strings.TrimSpace(text) // Remove leading/trailing whitespace
return text
}
type Orator interface { type Orator interface {
Speak(text string) error Speak(text string) error
Stop() Stop()
@@ -157,7 +121,7 @@ func (o *KokoroOrator) readroutine() {
} }
continue // if only one (often incomplete) sentence; wait for next chunk continue // if only one (often incomplete) sentence; wait for next chunk
} }
cleanedText := cleanText(sentence.Text) cleanedText := models.CleanText(sentence.Text)
if cleanedText == "" { if cleanedText == "" {
continue // Skip empty text after cleaning continue // Skip empty text after cleaning
} }
@@ -186,7 +150,7 @@ func (o *KokoroOrator) readroutine() {
// flush remaining text // flush remaining text
o.mu.Lock() o.mu.Lock()
remaining := o.textBuffer.String() remaining := o.textBuffer.String()
remaining = cleanText(remaining) remaining = models.CleanText(remaining)
o.textBuffer.Reset() o.textBuffer.Reset()
o.mu.Unlock() o.mu.Unlock()
if remaining == "" { if remaining == "" {
@@ -389,7 +353,7 @@ func (o *GoogleTranslateOrator) readroutine() {
} }
continue // if only one (often incomplete) sentence; wait for next chunk continue // if only one (often incomplete) sentence; wait for next chunk
} }
cleanedText := cleanText(sentence.Text) cleanedText := models.CleanText(sentence.Text)
if cleanedText == "" { if cleanedText == "" {
continue // Skip empty text after cleaning continue // Skip empty text after cleaning
} }
@@ -417,7 +381,7 @@ func (o *GoogleTranslateOrator) readroutine() {
} }
o.mu.Lock() o.mu.Lock()
remaining := o.textBuffer.String() remaining := o.textBuffer.String()
remaining = cleanText(remaining) remaining = models.CleanText(remaining)
o.textBuffer.Reset() o.textBuffer.Reset()
o.mu.Unlock() o.mu.Unlock()
if remaining == "" { if remaining == "" {

View File

@@ -5,6 +5,7 @@ import (
"gf-lt/models" "gf-lt/models"
"gf-lt/pngmeta" "gf-lt/pngmeta"
"image" "image"
"net/url"
"os" "os"
"path" "path"
"slices" "slices"
@@ -23,6 +24,25 @@ func isASCII(s string) bool {
return true return true
} }
// 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 {
if !cfg.StripThinkingFromAPI {
return msg
}
// Skip user, tool, and system messages - they might contain thinking examples
if msg.Role == cfg.UserRole || msg.Role == cfg.ToolRole || msg.Role == "system" {
return msg
}
// Strip thinking from assistant messages
if thinkRE.MatchString(msg.Content) {
msg.Content = thinkRE.ReplaceAllString(msg.Content, "")
// Clean up any double newlines that might result
msg.Content = strings.TrimSpace(msg.Content)
}
return msg
}
// refreshChatDisplay updates the chat display based on current character view // refreshChatDisplay updates the chat display based on current character view
// It filters messages for the character the user is currently "writing as" // It filters messages for the character the user is currently "writing as"
// and updates the textView with the filtered conversation // and updates the textView with the filtered conversation
@@ -35,14 +55,11 @@ func refreshChatDisplay() {
// Filter messages for this character // Filter messages for this character
filteredMessages := filterMessagesForCharacter(chatBody.Messages, viewingAs) filteredMessages := filterMessagesForCharacter(chatBody.Messages, viewingAs)
displayText := chatToText(filteredMessages, cfg.ShowSys) displayText := chatToText(filteredMessages, cfg.ShowSys)
// Use QueueUpdate for thread-safe UI updates textView.SetText(displayText)
app.QueueUpdate(func() { colorText()
textView.SetText(displayText) if scrollToEndEnabled {
colorText() textView.ScrollToEnd()
if scrollToEndEnabled { }
textView.ScrollToEnd()
}
})
} }
func stopTTSIfNotForUser(msg *models.RoleMsg) { func stopTTSIfNotForUser(msg *models.RoleMsg) {
@@ -243,6 +260,34 @@ func strInSlice(s string, sl []string) bool {
return false return false
} }
// isLocalLlamacpp checks if the current API is a local llama.cpp instance.
func isLocalLlamacpp() bool {
u, err := url.Parse(cfg.CurrentAPI)
if err != nil {
return false
}
host := u.Hostname()
return host == "localhost" || host == "127.0.0.1" || host == "::1"
}
// getModelColor returns the color tag for the model name based on its load status.
// For non-local models, returns orange. For local llama.cpp models, returns green if loaded, red if not.
func getModelColor() string {
if !isLocalLlamacpp() {
return "orange"
}
// Check if model is loaded
loaded, err := isModelLoaded(chatBody.Model)
if err != nil {
// On error, assume not loaded (red)
return "red"
}
if loaded {
return "green"
}
return "red"
}
func makeStatusLine() string { func makeStatusLine() string {
isRecording := false isRecording := false
if asr != nil { if asr != nil {
@@ -272,8 +317,10 @@ func makeStatusLine() string {
} else { } else {
shellModeInfo = "" shellModeInfo = ""
} }
// Get model color based on load status for local llama.cpp models
modelColor := getModelColor()
statusLine := fmt.Sprintf(indexLineCompletion, boolColors[botRespMode], botRespMode, activeChatName, statusLine := fmt.Sprintf(indexLineCompletion, boolColors[botRespMode], botRespMode, activeChatName,
boolColors[cfg.ToolUse], cfg.ToolUse, chatBody.Model, boolColors[cfg.SkipLLMResp], boolColors[cfg.ToolUse], cfg.ToolUse, modelColor, chatBody.Model, boolColors[cfg.SkipLLMResp],
cfg.SkipLLMResp, cfg.CurrentAPI, boolColors[isRecording], isRecording, persona, cfg.SkipLLMResp, cfg.CurrentAPI, boolColors[isRecording], isRecording, persona,
botPersona, boolColors[injectRole], injectRole) botPersona, boolColors[injectRole], injectRole)
return statusLine + imageInfo + shellModeInfo return statusLine + imageInfo + shellModeInfo

82
llm.go
View File

@@ -13,28 +13,6 @@ var imageAttachmentPath string // Global variable to track image attachment for
var lastImg string // for ctrl+j var lastImg string // for ctrl+j
var RAGMsg = "Retrieved context for user's query:\n" var RAGMsg = "Retrieved context for user's query:\n"
// addPersonaSuffixToLastUserMessage adds the persona suffix to the last user message
// to indicate to the assistant who it should reply as
func addPersonaSuffixToLastUserMessage(messages []models.RoleMsg, persona string) []models.RoleMsg {
if len(messages) == 0 {
return messages
}
// // Find the last user message to modify
// for i := len(messages) - 1; i >= 0; i-- {
// if messages[i].Role == cfg.UserRole || messages[i].Role == "user" {
// // Create a copy of the message to avoid modifying the original
// modifiedMsg := messages[i]
// modifiedMsg.Content = modifiedMsg.Content + "\n" + persona + ":"
// messages[i] = modifiedMsg
// break
// }
// }
modifiedMsg := messages[len(messages)-1]
modifiedMsg.Content = modifiedMsg.Content + "\n" + persona + ":\n"
messages[len(messages)-1] = modifiedMsg
return messages
}
// containsToolSysMsg checks if the toolSysMsg already exists in the chat body // containsToolSysMsg checks if the toolSysMsg already exists in the chat body
func containsToolSysMsg() bool { func containsToolSysMsg() bool {
for _, msg := range chatBody.Messages { for _, msg := range chatBody.Messages {
@@ -187,17 +165,9 @@ func (lcp LCPCompletion) FormMsg(msg, role string, resume bool) (io.Reader, erro
filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages) filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages)
messages := make([]string, len(filteredMessages)) messages := make([]string, len(filteredMessages))
for i, m := range filteredMessages { for i, m := range filteredMessages {
messages[i] = m.ToPrompt() messages[i] = stripThinkingFromMsg(&m).ToPrompt()
} }
prompt := strings.Join(messages, "\n") prompt := strings.Join(messages, "\n")
// strings builder?
if !resume {
botMsgStart := "\n" + botPersona + ":\n"
prompt += botMsgStart
}
if cfg.ThinkUse && !cfg.ToolUse {
prompt += "<think>"
}
// Add multimodal media markers to the prompt text when multimodal data is present // 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 // This is required by llama.cpp multimodal models so they know where to insert media
if len(multimodalData) > 0 { if len(multimodalData) > 0 {
@@ -209,6 +179,14 @@ func (lcp LCPCompletion) FormMsg(msg, role string, resume bool) (io.Reader, erro
} }
prompt = sb.String() prompt = sb.String()
} }
// needs to be after <__media__> if there are images
if !resume {
botMsgStart := "\n" + botPersona + ":\n"
prompt += botMsgStart
}
if cfg.ThinkUse && !cfg.ToolUse {
prompt += "<think>"
}
logger.Debug("checking prompt for /completion", "tool_use", cfg.ToolUse, logger.Debug("checking prompt for /completion", "tool_use", cfg.ToolUse,
"msg", msg, "resume", resume, "prompt", prompt, "multimodal_data_count", len(multimodalData)) "msg", msg, "resume", resume, "prompt", prompt, "multimodal_data_count", len(multimodalData))
payload := models.NewLCPReq(prompt, chatBody.Model, multimodalData, payload := models.NewLCPReq(prompt, chatBody.Model, multimodalData,
@@ -311,7 +289,7 @@ func (op LCPChat) FormMsg(msg, role string, resume bool) (io.Reader, error) {
// If image processing fails, fall back to simple text message // If image processing fails, fall back to simple text message
newMsg = models.NewRoleMsg(role, msg) newMsg = models.NewRoleMsg(role, msg)
} else { } else {
newMsg.AddImagePart(imageURL) newMsg.AddImagePart(imageURL, localImageAttachmentPath)
} }
// Only clear the global image attachment after successfully processing it in this API call // Only clear the global image attachment after successfully processing it in this API call
imageAttachmentPath = "" // Clear the attachment after use imageAttachmentPath = "" // Clear the attachment after use
@@ -341,23 +319,21 @@ func (op LCPChat) FormMsg(msg, role string, resume bool) (io.Reader, error) {
logger.Debug("LCPChat: RAG message added to chat body", "role", ragMsg.Role, logger.Debug("LCPChat: RAG message added to chat body", "role", ragMsg.Role,
"rag_content_len", len(ragMsg.Content), "message_count_after_rag", len(chatBody.Messages)) "rag_content_len", len(ragMsg.Content), "message_count_after_rag", len(chatBody.Messages))
} }
filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages) filteredMessages, _ := filterMessagesForCurrentCharacter(chatBody.Messages)
// openai /v1/chat does not support custom roles; needs to be user, assistant, system // openai /v1/chat does not support custom roles; needs to be user, assistant, system
// Add persona suffix to the last user message to indicate who the assistant should reply as // Add persona suffix to the last user message to indicate who the assistant should reply as
if cfg.AutoTurn && !resume {
filteredMessages = addPersonaSuffixToLastUserMessage(filteredMessages, botPersona)
}
bodyCopy := &models.ChatBody{ bodyCopy := &models.ChatBody{
Messages: make([]models.RoleMsg, len(filteredMessages)), Messages: make([]models.RoleMsg, len(filteredMessages)),
Model: chatBody.Model, Model: chatBody.Model,
Stream: chatBody.Stream, Stream: chatBody.Stream,
} }
for i, msg := range filteredMessages { for i, msg := range filteredMessages {
if msg.Role == cfg.UserRole { strippedMsg := *stripThinkingFromMsg(&msg)
bodyCopy.Messages[i] = msg if strippedMsg.Role == cfg.UserRole {
bodyCopy.Messages[i] = strippedMsg
bodyCopy.Messages[i].Role = "user" bodyCopy.Messages[i].Role = "user"
} else { } else {
bodyCopy.Messages[i] = msg bodyCopy.Messages[i] = strippedMsg
} }
} }
// Clean null/empty messages to prevent API issues // Clean null/empty messages to prevent API issues
@@ -437,7 +413,7 @@ func (ds DeepSeekerCompletion) FormMsg(msg, role string, resume bool) (io.Reader
filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages) filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages)
messages := make([]string, len(filteredMessages)) messages := make([]string, len(filteredMessages))
for i, m := range filteredMessages { for i, m := range filteredMessages {
messages[i] = m.ToPrompt() messages[i] = stripThinkingFromMsg(&m).ToPrompt()
} }
prompt := strings.Join(messages, "\n") prompt := strings.Join(messages, "\n")
// strings builder? // strings builder?
@@ -519,22 +495,20 @@ func (ds DeepSeekerChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages)) logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages))
} }
// Create copy of chat body with standardized user role // Create copy of chat body with standardized user role
filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages) filteredMessages, _ := filterMessagesForCurrentCharacter(chatBody.Messages)
// Add persona suffix to the last user message to indicate who the assistant should reply as // Add persona suffix to the last user message to indicate who the assistant should reply as
if cfg.AutoTurn && !resume {
filteredMessages = addPersonaSuffixToLastUserMessage(filteredMessages, botPersona)
}
bodyCopy := &models.ChatBody{ bodyCopy := &models.ChatBody{
Messages: make([]models.RoleMsg, len(filteredMessages)), Messages: make([]models.RoleMsg, len(filteredMessages)),
Model: chatBody.Model, Model: chatBody.Model,
Stream: chatBody.Stream, Stream: chatBody.Stream,
} }
for i, msg := range filteredMessages { for i, msg := range filteredMessages {
if msg.Role == cfg.UserRole || i == 1 { strippedMsg := *stripThinkingFromMsg(&msg)
bodyCopy.Messages[i] = msg if strippedMsg.Role == cfg.UserRole || i == 1 {
bodyCopy.Messages[i] = strippedMsg
bodyCopy.Messages[i].Role = "user" bodyCopy.Messages[i].Role = "user"
} else { } else {
bodyCopy.Messages[i] = msg bodyCopy.Messages[i] = strippedMsg
} }
} }
// Clean null/empty messages to prevent API issues // Clean null/empty messages to prevent API issues
@@ -605,7 +579,7 @@ func (or OpenRouterCompletion) FormMsg(msg, role string, resume bool) (io.Reader
filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages) filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages)
messages := make([]string, len(filteredMessages)) messages := make([]string, len(filteredMessages))
for i, m := range filteredMessages { for i, m := range filteredMessages {
messages[i] = m.ToPrompt() messages[i] = stripThinkingFromMsg(&m).ToPrompt()
} }
prompt := strings.Join(messages, "\n") prompt := strings.Join(messages, "\n")
// strings builder? // strings builder?
@@ -690,7 +664,7 @@ func (or OpenRouterChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
// If image processing fails, fall back to simple text message // If image processing fails, fall back to simple text message
newMsg = models.NewRoleMsg(role, msg) newMsg = models.NewRoleMsg(role, msg)
} else { } else {
newMsg.AddImagePart(imageURL) newMsg.AddImagePart(imageURL, localImageAttachmentPath)
} }
// Only clear the global image attachment after successfully processing it in this API call // Only clear the global image attachment after successfully processing it in this API call
imageAttachmentPath = "" // Clear the attachment after use imageAttachmentPath = "" // Clear the attachment after use
@@ -718,21 +692,19 @@ func (or OpenRouterChat) FormMsg(msg, role string, resume bool) (io.Reader, erro
logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages)) logger.Debug("RAG message added to chat body", "message_count", len(chatBody.Messages))
} }
// Create copy of chat body with standardized user role // Create copy of chat body with standardized user role
filteredMessages, botPersona := filterMessagesForCurrentCharacter(chatBody.Messages) filteredMessages, _ := filterMessagesForCurrentCharacter(chatBody.Messages)
// Add persona suffix to the last user message to indicate who the assistant should reply as // Add persona suffix to the last user message to indicate who the assistant should reply as
if cfg.AutoTurn && !resume {
filteredMessages = addPersonaSuffixToLastUserMessage(filteredMessages, botPersona)
}
bodyCopy := &models.ChatBody{ bodyCopy := &models.ChatBody{
Messages: make([]models.RoleMsg, len(filteredMessages)), Messages: make([]models.RoleMsg, len(filteredMessages)),
Model: chatBody.Model, Model: chatBody.Model,
Stream: chatBody.Stream, Stream: chatBody.Stream,
} }
for i, msg := range filteredMessages { for i, msg := range filteredMessages {
bodyCopy.Messages[i] = msg strippedMsg := *stripThinkingFromMsg(&msg)
bodyCopy.Messages[i] = strippedMsg
// Standardize role if it's a user role // Standardize role if it's a user role
if bodyCopy.Messages[i].Role == cfg.UserRole { if bodyCopy.Messages[i].Role == cfg.UserRole {
bodyCopy.Messages[i] = msg bodyCopy.Messages[i] = strippedMsg
bodyCopy.Messages[i].Role = "user" bodyCopy.Messages[i].Role = "user"
} }
} }

View File

@@ -12,7 +12,8 @@ var (
injectRole = true injectRole = true
selectedIndex = int(-1) selectedIndex = int(-1)
shellMode = false shellMode = false
indexLineCompletion = "F12 to show keys help | llm turn: [%s:-:b]%v[-:-:-] (F6) | chat: [orange:-:b]%s[-:-:-] (F1) | toolUseAdviced: [%s:-:b]%v[-:-:-] (ctrl+k) | model: [orange:-:b]%s[-:-:-] (ctrl+l) | skip LLM resp: [%s:-:b]%v[-:-:-] (F10)\nAPI: [orange:-:b]%s[-:-:-] (ctrl+v) | recording: [%s:-:b]%v[-:-:-] (ctrl+r) | writing as: [orange:-:b]%s[-:-:-] (ctrl+q) | bot will write as [orange:-:b]%s[-:-:-] (ctrl+x) | role injection (alt+7) [%s:-:b]%v[-:-:-]" thinkingCollapsed = false
indexLineCompletion = "F12 to show keys help | llm turn: [%s:-:b]%v[-:-:-] (F6) | chat: [orange:-:b]%s[-:-:-] (F1) | toolUseAdviced: [%s:-:b]%v[-:-:-] (ctrl+k) | model: [%s:-:b]%s[-:-:-] (ctrl+l) | skip LLM resp: [%s:-:b]%v[-:-:-] (F10)\nAPI: [orange:-:b]%s[-:-:-] (ctrl+v) | recording: [%s:-:b]%v[-:-:-] (ctrl+r) | writing as: [orange:-:b]%s[-:-:-] (ctrl+q) | bot will write as [orange:-:b]%s[-:-:-] (ctrl+x) | role injection (alt+7) [%s:-:b]%v[-:-:-]"
focusSwitcher = map[tview.Primitive]tview.Primitive{} focusSwitcher = map[tview.Primitive]tview.Primitive{}
) )

View File

@@ -1,8 +1,49 @@
package models package models
import (
"regexp"
"strings"
)
type AudioFormat string type AudioFormat string
const ( const (
AFWav AudioFormat = "wav" AFWav AudioFormat = "wav"
AFMP3 AudioFormat = "mp3" AFMP3 AudioFormat = "mp3"
) )
var threeOrMoreDashesRE = regexp.MustCompile(`-{3,}`)
// CleanText removes markdown and special characters that are not suitable for TTS
func CleanText(text string) string {
// Remove markdown-like characters that might interfere with TTS
text = strings.ReplaceAll(text, "*", "") // Bold/italic markers
text = strings.ReplaceAll(text, "#", "") // Headers
text = strings.ReplaceAll(text, "_", "") // Underline/italic markers
text = strings.ReplaceAll(text, "~", "") // Strikethrough markers
text = strings.ReplaceAll(text, "`", "") // Code markers
text = strings.ReplaceAll(text, "[", "") // Link brackets
text = strings.ReplaceAll(text, "]", "") // Link brackets
text = strings.ReplaceAll(text, "!", "") // Exclamation marks (if not punctuation)
// Remove HTML tags using regex
htmlTagRegex := regexp.MustCompile(`<[^>]*>`)
text = htmlTagRegex.ReplaceAllString(text, "")
// Split text into lines to handle table separators
lines := strings.Split(text, "\n")
var filteredLines []string
for _, line := range lines {
// Check if the line looks like a table separator (e.g., |----|, |===|, | - - - |)
// A table separator typically contains only |, -, =, and spaces
isTableSeparator := regexp.MustCompile(`^\s*\|\s*[-=\s]+\|\s*$`).MatchString(strings.TrimSpace(line))
if !isTableSeparator {
// If it's not a table separator, remove vertical bars but keep the content
processedLine := strings.ReplaceAll(line, "|", "")
filteredLines = append(filteredLines, processedLine)
}
// If it is a table separator, skip it (don't add to filteredLines)
}
text = strings.Join(filteredLines, "\n")
text = threeOrMoreDashesRE.ReplaceAllString(text, "")
text = strings.TrimSpace(text) // Remove leading/trailing whitespace
return text
}

View File

@@ -5,9 +5,22 @@ import (
"encoding/json" "encoding/json"
"fmt" "fmt"
"os" "os"
"path/filepath"
"strings" "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 { type FuncCall struct {
ID string `json:"id,omitempty"` ID string `json:"id,omitempty"`
Name string `json:"name"` Name string `json:"name"`
@@ -82,6 +95,7 @@ type TextContentPart struct {
type ImageContentPart struct { type ImageContentPart struct {
Type string `json:"type"` Type string `json:"type"`
Path string `json:"path,omitempty"` // Store original file path
ImageURL struct { ImageURL struct {
URL string `json:"url"` URL string `json:"url"`
} `json:"image_url"` } `json:"image_url"`
@@ -169,10 +183,11 @@ func (m *RoleMsg) UnmarshalJSON(data []byte) error {
func (m *RoleMsg) ToText(i int) string { func (m *RoleMsg) ToText(i int) string {
// Convert content to string representation // Convert content to string representation
var contentStr string var contentStr string
var imageIndicators []string
if !m.hasContentParts { if !m.hasContentParts {
contentStr = m.Content contentStr = m.Content
} else { } else {
// For structured content, just take the text parts // For structured content, collect text parts and image indicators
var textParts []string var textParts []string
for _, part := range m.ContentParts { for _, part := range m.ContentParts {
switch p := part.(type) { switch p := part.(type) {
@@ -181,13 +196,35 @@ func (m *RoleMsg) ToText(i int) string {
textParts = append(textParts, p.Text) textParts = append(textParts, p.Text)
} }
case ImageContentPart: case ImageContentPart:
// skip images for text display // Collect image indicator
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: case map[string]any:
if partType, exists := p["type"]; exists && partType == "text" { if partType, exists := p["type"]; exists {
if textVal, textExists := p["text"]; textExists { switch partType {
if textStr, isStr := textVal.(string); isStr { case "text":
textParts = append(textParts, textStr) if textVal, textExists := p["text"]; textExists {
if textStr, isStr := textVal.(string); isStr {
textParts = append(textParts, textStr)
}
} }
case "image_url":
// Handle unmarshaled image content
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))
} }
} }
} }
@@ -201,7 +238,17 @@ func (m *RoleMsg) ToText(i int) string {
// if !strings.HasPrefix(contentStr, m.Role+":") { // if !strings.HasPrefix(contentStr, m.Role+":") {
icon := fmt.Sprintf("(%d) <%s>: ", i, m.Role) icon := fmt.Sprintf("(%d) <%s>: ", i, m.Role)
// } // }
textMsg := fmt.Sprintf("[-:-:b]%s[-:-:-]\n%s\n", icon, contentStr) // Build final message with image indicators before text
var finalContent strings.Builder
if len(imageIndicators) > 0 {
// Add each image indicator on its own line
for _, indicator := range imageIndicators {
finalContent.WriteString(indicator)
finalContent.WriteString("\n")
}
}
finalContent.WriteString(contentStr)
textMsg := fmt.Sprintf("[-:-:b]%s[-:-:-]\n%s\n", icon, finalContent.String())
return strings.ReplaceAll(textMsg, "\n\n", "\n") return strings.ReplaceAll(textMsg, "\n\n", "\n")
} }
@@ -303,7 +350,7 @@ func (m *RoleMsg) AddTextPart(text string) {
} }
// AddImagePart adds an image content part to the message // AddImagePart adds an image content part to the message
func (m *RoleMsg) AddImagePart(imageURL string) { func (m *RoleMsg) AddImagePart(imageURL, imagePath string) {
if !m.hasContentParts { if !m.hasContentParts {
// Convert to content parts format // Convert to content parts format
if m.Content != "" { if m.Content != "" {
@@ -316,6 +363,7 @@ func (m *RoleMsg) AddImagePart(imageURL string) {
imagePart := ImageContentPart{ imagePart := ImageContentPart{
Type: "image_url", Type: "image_url",
Path: imagePath, // Store the original file path
ImageURL: struct { ImageURL: struct {
URL string `json:"url"` URL string `json:"url"`
}{URL: imageURL}, }{URL: imageURL},
@@ -355,6 +403,31 @@ func CreateImageURLFromPath(imagePath string) (string, error) {
return fmt.Sprintf("data:%s;base64,%s", mimeType, encoded), nil 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 { type ChatBody struct {
Model string `json:"model"` Model string `json:"model"`
Stream bool `json:"stream"` Stream bool `json:"stream"`

167
models/models_test.go Normal file
View File

@@ -0,0 +1,167 @@
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)
}
})
}
}

View File

@@ -928,11 +928,12 @@ func makeFilePicker() *tview.Flex {
} }
} }
// Update status line based on search state // Update status line based on search state
if searching { switch {
case searching:
statusView.SetText("Search: " + searchQuery + "_") statusView.SetText("Search: " + searchQuery + "_")
} else if searchQuery != "" { case searchQuery != "":
statusView.SetText("Current: " + dir + " (filter: " + searchQuery + ")") statusView.SetText("Current: " + dir + " (filter: " + searchQuery + ")")
} else { default:
statusView.SetText("Current: " + dir) statusView.SetText("Current: " + dir)
} }
} }

28
tui.go
View File

@@ -83,6 +83,7 @@ var (
[yellow]Ctrl+l[white]: show model selection popup to choose current model [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+k[white]: switch tool use (recommend tool use to llm after user msg)
[yellow]Ctrl+a[white]: interrupt tts (needs tts server) [yellow]Ctrl+a[white]: interrupt tts (needs tts server)
[yellow]Alt+0[white]: replay last message via tts (needs tts server)
[yellow]Ctrl+g[white]: open RAG file manager (load files for context retrieval) [yellow]Ctrl+g[white]: open RAG file manager (load files for context retrieval)
[yellow]Ctrl+y[white]: list loaded RAG files (view and manage loaded files) [yellow]Ctrl+y[white]: list loaded RAG files (view and manage loaded files)
[yellow]Ctrl+q[white]: show user role selection popup to choose who sends next msg as [yellow]Ctrl+q[white]: show user role selection popup to choose who sends next msg as
@@ -96,6 +97,7 @@ var (
[yellow]Alt+7[white]: toggle role injection (inject role in messages) [yellow]Alt+7[white]: toggle role injection (inject role in messages)
[yellow]Alt+8[white]: show char img or last picked img [yellow]Alt+8[white]: show char img or last picked img
[yellow]Alt+9[white]: warm up (load) selected llama.cpp model [yellow]Alt+9[white]: warm up (load) selected llama.cpp model
[yellow]Alt+t[white]: toggle thinking blocks visibility (collapse/expand <think> blocks)
=== scrolling chat window (some keys similar to vim) === === scrolling chat window (some keys similar to vim) ===
[yellow]arrows up/down and j/k[white]: scroll up and down [yellow]arrows up/down and j/k[white]: scroll up and down
@@ -831,6 +833,20 @@ func init() {
injectRole = !injectRole injectRole = !injectRole
updateStatusLine() updateStatusLine()
} }
// Handle Alt+T to toggle thinking block visibility
if event.Key() == tcell.KeyRune && event.Rune() == 't' && event.Modifiers()&tcell.ModAlt != 0 {
thinkingCollapsed = !thinkingCollapsed
textView.SetText(chatToText(chatBody.Messages, cfg.ShowSys))
colorText()
status := "expanded"
if thinkingCollapsed {
status = "collapsed"
}
if err := notifyUser("thinking", "Thinking blocks "+status); err != nil {
logger.Error("failed to send notification", "error", err)
}
return nil
}
if event.Key() == tcell.KeyF1 { if event.Key() == tcell.KeyF1 {
// chatList, err := loadHistoryChats() // chatList, err := loadHistoryChats()
chatList, err := store.GetChatByChar(cfg.AssistantRole) chatList, err := store.GetChatByChar(cfg.AssistantRole)
@@ -1129,6 +1145,18 @@ func init() {
if event.Key() == tcell.KeyCtrlA && cfg.TTS_ENABLED { if event.Key() == tcell.KeyCtrlA && cfg.TTS_ENABLED {
TTSDoneChan <- true TTSDoneChan <- true
} }
if event.Key() == tcell.KeyRune && event.Rune() == '0' && event.Modifiers()&tcell.ModAlt != 0 && cfg.TTS_ENABLED {
if len(chatBody.Messages) > 0 {
// Stop any currently playing TTS first
TTSDoneChan <- true
lastMsg := chatBody.Messages[len(chatBody.Messages)-1]
cleanedText := models.CleanText(lastMsg.Content)
if cleanedText != "" {
go orator.Speak(cleanedText)
}
}
return nil
}
if event.Key() == tcell.KeyCtrlW { if event.Key() == tcell.KeyCtrlW {
// INFO: continue bot/text message // INFO: continue bot/text message
// without new role // without new role