feat: add --thinking flag for model reasoning depth

Supports Deepseek (enabled/disabled → thinking.type) and
OpenAI o-series (low/medium/high → reasoning_effort).
Wired through Config → Client → CallParams → CLI flag.

Author: molty3000

cmd/kode/main.go

@@ -45,13 +45,14 @@ Flags:
   --model <name>       LLM model (default: deepseek-chat)
   --base-url <url>     API endpoint (default: https://api.deepseek.com/v1)
   --max-iter <n>       Max think->act cycles (default: 90)
+  --thinking <level>   Reasoning depth: enabled|disabled (Deepseek) or low|medium|high (OpenAI o-series)
   --sandbox            Run in isolated Docker container
   --system <prompt>    System prompt override`)
 }
 
 func runCmd() {
 	args := os.Args[2:]
-	var model, baseURL, system string
+	var model, baseURL, system, thinking string
 	maxIter := 90
 	sandbox := false
 
@@ -70,6 +71,9 @@ func runCmd() {
 		case "--system":
 			system = args[i+1]
 			i += 2
+		case "--thinking":
+			thinking = args[i+1]
+			i += 2
 		case "--sandbox":
 			sandbox = true
 			i++
@@ -97,6 +101,7 @@ done:
 		BaseURL:       baseURL,
 		MaxIterations: maxIter,
 		SystemMessage: system,
+		Thinking:      thinking,
 		Tools:         builtinTools(),
 	})
 	if err != nil {

internal/llm/client.go

@@ -14,19 +14,21 @@ import (
 
 // Client sends chat completion requests to any OpenAI-compatible endpoint.
 type Client struct {
-	BaseURL string
-	APIKey  string
-	Model   string
-	http    *http.Client
+	BaseURL  string
+	APIKey   string
+	Model    string
+	Thinking string // "enabled", "disabled", "low", "medium", "high", or empty
+	http     *http.Client
 }
 
 // New creates a Client with sensible defaults.
-func New(baseURL, apiKey, model string) *Client {
+func New(baseURL, apiKey, model, thinking string) *Client {
 	return &Client{
-		BaseURL: strings.TrimRight(baseURL, "/"),
-		APIKey:  apiKey,
-		Model:   model,
-		http:    &http.Client{Timeout: 120 * time.Second},
+		BaseURL:  strings.TrimRight(baseURL, "/"),
+		APIKey:   apiKey,
+		Model:    model,
+		Thinking: thinking,
+		http:     &http.Client{Timeout: 120 * time.Second},
 	}
 }
 
@@ -65,10 +67,17 @@ type FunctionDef struct {
 
 // CallParams is the request body for /chat/completions.
 type CallParams struct {
-	Model    string    `json:"model"`
-	Messages []Message `json:"messages"`
-	Tools    []ToolDef `json:"tools,omitempty"`
-	Stream   bool      `json:"stream"`
+	Model          string          `json:"model"`
+	Messages       []Message       `json:"messages"`
+	Tools          []ToolDef       `json:"tools,omitempty"`
+	Stream         bool            `json:"stream"`
+	Thinking       *ThinkingConfig `json:"thinking,omitempty"`
+	ReasoningEffort string         `json:"reasoning_effort,omitempty"`
+}
+
+// ThinkingConfig controls Deepseek's extended thinking feature.
+type ThinkingConfig struct {
+	Type string `json:"type"` // "enabled" or "disabled"
 }
 
 // CallResult is the parsed response from /chat/completions.
@@ -89,6 +98,13 @@ func (c *Client) Call(ctx context.Context, messages []Message, tools []ToolDef)
 		Stream:   false,
 	}
 
+	switch c.Thinking {
+	case "enabled", "disabled":
+		body.Thinking = &ThinkingConfig{Type: c.Thinking}
+	case "low", "medium", "high":
+		body.ReasoningEffort = c.Thinking
+	}
+
 	reqBytes, err := json.Marshal(body)
 	if err != nil {
 		return nil, fmt.Errorf("llm: marshal request: %w", err)

kode.go

@@ -50,6 +50,14 @@ type Config struct {
 	// Falls back to DEEPSEEK_API_KEY, then OPENAI_API_KEY env vars.
 	APIKey string
 
+	// Thinking controls the model's reasoning depth. Provider-specific:
+	//
+	//   Deepseek: "enabled" or "disabled" → {"type": "enabled"}
+	//   OpenAI o-series: "low", "medium", "high" → {"reasoning_effort": "low"}
+	//
+	// Leave empty for provider default behavior.
+	Thinking string
+
 	// Tools available to the agent.
 	Tools []Tool
 
@@ -95,7 +103,7 @@ func New(cfg Config) (*Agent, error) {
 	}
 
 	registry := tool.NewRegistry(tools)
-	client := llm.New(cfg.BaseURL, cfg.APIKey, cfg.Model)
+	client := llm.New(cfg.BaseURL, cfg.APIKey, cfg.Model, cfg.Thinking)
 	engine := loop.New(client, registry, cfg.MaxIterations, cfg.SystemMessage)
 
 	return &Agent{