feat(lsp): add coding_temperature knob and remove hardcoded temps\n\n- Add to app config and server options.\n- Use in LSP code actions and completions.\n- Default to provider temperature when not set.\n- Update README and config.json.example.

6 files changed, 51 insertions, 22 deletions

diff --git a/README.md b/README.md
index cb74f95..b8a85a4 100644
--- a/README.md
+++ b/README.md
@@ -24,6 +24,7 @@ Hexai exposes a simple LLM provider interface. It supports OpenAI, GitHub Copilo
   "log_preview_limit": 100,
   "no_disk_io": true,
   "trigger_characters": [".", ":", "/", "_", ";", "?"],
+  "coding_temperature": 0.2,
   "provider": "ollama",
   "copilot_model": "gpt-4.1",
   "copilot_base_url": "https://api.githubcopilot.com",
@@ -39,6 +40,7 @@ Hexai exposes a simple LLM provider interface. It supports OpenAI, GitHub Copilo
 
 * context_mode: minimal | window | file-on-new-func | always-full
 * provider: openai | copilot | ollama
+* coding_temperature: single knob for LSP requests (optional; default uses provider temperature)
 * openai_model, openai_base_url, openai_temperature: OpenAI-only options
 * copilot_model, copilot_base_url, copilot_temperature: Copilot-only options
 * ollama_model, ollama_base_url, ollama_temperature: Ollama-only options
@@ -84,8 +86,9 @@ Ensure `OPENAI_API_KEY` or `COPILOT_API_KEY` is set in your environment accordin
   `ollama_temperature` to override. Valid ranges depend on the provider, but
   typically `0.0`–`2.0`.
 * LSP vs CLI: The LSP sometimes overrides temperature for specific actions
-  (e.g., `0.1`–`0.2` for completions). The CLI uses the configured provider
-  default unless you change it.
+  using `coding_temperature` (if set). If `coding_temperature` is not set,
+  LSP calls use the provider default temperature. The CLI uses the configured
+  provider default unless you change it.
 
 Recommended ranges and use cases:
 
diff --git a/config.json.example b/config.json.example
index 04ecfb7..b8c729b 100644
--- a/config.json.example
+++ b/config.json.example
@@ -6,6 +6,7 @@
   "log_preview_limit": 100,
   "no_disk_io": true,
   "trigger_characters": [".", ":", "/", "_", ";", "?"],
+  "coding_temperature": 0.2,
 
   "provider": "openai",
 
diff --git a/internal/appconfig/config.go b/internal/appconfig/config.go
index c377467..4fa3441 100644
--- a/internal/appconfig/config.go
+++ b/internal/appconfig/config.go
@@ -17,7 +17,9 @@ type App struct {
 	ContextMode        string `json:"context_mode"`
 	ContextWindowLines int    `json:"context_window_lines"`
 	MaxContextTokens   int    `json:"max_context_tokens"`
-	LogPreviewLimit    int    `json:"log_preview_limit"`
+    LogPreviewLimit    int    `json:"log_preview_limit"`
+    // Single knob for LSP requests; if set, overrides hardcoded temps in LSP.
+    CodingTemperature  *float64 `json:"coding_temperature"`
 
 	TriggerCharacters []string `json:"trigger_characters"`
 	Provider          string   `json:"provider"`
@@ -48,6 +50,7 @@ func newDefaultConfig() App {
         ContextWindowLines: 120,
         MaxContextTokens:   4000,
         LogPreviewLimit:    100,
+        CodingTemperature:  &t,
         OpenAITemperature:  &t,
         OllamaTemperature:  &t,
         CopilotTemperature: &t,
@@ -115,6 +118,9 @@ func (a *App) mergeWith(other *App) {
     if other.LogPreviewLimit >= 0 {
         a.LogPreviewLimit = other.LogPreviewLimit
     }
+    if other.CodingTemperature != nil { // allow explicit 0.0
+        a.CodingTemperature = other.CodingTemperature
+    }
     if len(other.TriggerCharacters) > 0 {
         a.TriggerCharacters = slices.Clone(other.TriggerCharacters)
     }
diff --git a/internal/hexailsp/run.go b/internal/hexailsp/run.go
index 2eee3aa..1beb93a 100644
--- a/internal/hexailsp/run.go
+++ b/internal/hexailsp/run.go
@@ -83,6 +83,8 @@ func RunWithFactory(logPath string, stdin io.Reader, stdout io.Writer, logger *l
 		WindowLines:      cfg.ContextWindowLines,
 		MaxContextTokens: cfg.MaxContextTokens,
 
+		CodingTemperature: cfg.CodingTemperature,
+
 		Client:            client,
 		TriggerCharacters: cfg.TriggerCharacters,
 	})
diff --git a/internal/lsp/handlers.go b/internal/lsp/handlers.go
index 7903c73..d21c5b3 100644
--- a/internal/lsp/handlers.go
+++ b/internal/lsp/handlers.go
@@ -93,7 +93,12 @@ func (s *Server) handleCodeAction(req Request) {
 		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
 		defer cancel()
 		messages := []llm.Message{{Role: "system", Content: sys}, {Role: "user", Content: user}}
-		if text, err := s.llmClient.Chat(ctx, messages, llm.WithMaxTokens(s.maxTokens), llm.WithTemperature(0.1)); err == nil {
+        // Build request options from server settings
+        opts := []llm.RequestOption{llm.WithMaxTokens(s.maxTokens)}
+        if s.codingTemperature != nil {
+            opts = append(opts, llm.WithTemperature(*s.codingTemperature))
+        }
+        if text, err := s.llmClient.Chat(ctx, messages, opts...); err == nil {
 			out := strings.TrimSpace(text)
 			if out != "" {
 				edit := WorkspaceEdit{Changes: map[string][]TextEdit{p.TextDocument.URI: {{Range: p.Range, NewText: out}}}}
@@ -123,7 +128,11 @@ func (s *Server) handleCodeAction(req Request) {
 		ctx, cancel := context.WithTimeout(context.Background(), 12*time.Second)
 		defer cancel()
 		messages := []llm.Message{{Role: "system", Content: sys}, {Role: "user", Content: b.String()}}
-		if text, err := s.llmClient.Chat(ctx, messages, llm.WithMaxTokens(s.maxTokens), llm.WithTemperature(0.1)); err == nil {
+        opts := []llm.RequestOption{llm.WithMaxTokens(s.maxTokens)}
+        if s.codingTemperature != nil {
+            opts = append(opts, llm.WithTemperature(*s.codingTemperature))
+        }
+        if text, err := s.llmClient.Chat(ctx, messages, opts...); err == nil {
 			out := strings.TrimSpace(text)
 			if out != "" {
 				edit := WorkspaceEdit{Changes: map[string][]TextEdit{p.TextDocument.URI: {{Range: p.Range, NewText: out}}}}
@@ -454,7 +463,11 @@ func (s *Server) tryLLMCompletion(p CompletionParams, above, current, below, fun
 	s.llmSentBytesTotal += int64(sentSize)
 	s.mu.Unlock()
 
-	text, err := s.llmClient.Chat(ctx, messages, llm.WithMaxTokens(s.maxTokens), llm.WithTemperature(0.2))
+    opts := []llm.RequestOption{llm.WithMaxTokens(s.maxTokens)}
+    if s.codingTemperature != nil {
+        opts = append(opts, llm.WithTemperature(*s.codingTemperature))
+    }
+    text, err := s.llmClient.Chat(ctx, messages, opts...)
 	if err != nil {
 		logging.Logf("lsp ", "llm completion error: %v", err)
 		// Log updated averages after this request (even if failed)
diff --git a/internal/lsp/server.go b/internal/lsp/server.go
index 5f4423a..f1ca302 100644
--- a/internal/lsp/server.go
+++ b/internal/lsp/server.go
@@ -29,7 +29,9 @@ type Server struct {
 	windowLines      int
 	maxContextTokens int
 	noDiskIO         bool
-	triggerChars     []string
+    triggerChars     []string
+    // If set, used as the LSP coding temperature for all LLM calls
+    codingTemperature *float64
 	// LLM request stats
 	llmReqTotal       int64
 	llmSentBytesTotal int64
@@ -40,14 +42,15 @@ type Server struct {
 
 // ServerOptions collects configuration for NewServer to avoid long parameter lists.
 type ServerOptions struct {
-	LogContext        bool
-	MaxTokens         int
-	ContextMode       string
-	WindowLines       int
-	MaxContextTokens  int
-	
-	Client            llm.Client
-	TriggerCharacters []string
+    LogContext        bool
+    MaxTokens         int
+    ContextMode       string
+    WindowLines       int
+    MaxContextTokens  int
+
+    Client            llm.Client
+    TriggerCharacters []string
+    CodingTemperature *float64
 }
 
 func NewServer(r io.Reader, w io.Writer, logger *log.Logger, opts ServerOptions) *Server {
@@ -74,13 +77,14 @@ func NewServer(r io.Reader, w io.Writer, logger *log.Logger, opts ServerOptions)
 	s.maxContextTokens = maxContextTokens
 	
 	s.startTime = time.Now()
-	s.llmClient = opts.Client
-	if len(opts.TriggerCharacters) == 0 {
-		s.triggerChars = []string{".", ":", "/", "_", ";", "?"}
-	} else {
-		s.triggerChars = append([]string{}, opts.TriggerCharacters...)
-	}
-	return s
+    s.llmClient = opts.Client
+    if len(opts.TriggerCharacters) == 0 {
+        s.triggerChars = []string{".", ":", "/", "_", ";", "?"}
+    } else {
+        s.triggerChars = append([]string{}, opts.TriggerCharacters...)
+    }
+    s.codingTemperature = opts.CodingTemperature
+    return s
 }
 
 func (s *Server) Run() error {