quick toggling of thinking, session wide

can be set from the api as well, just like "yolo"

Signed-off-by: Christopher Petito <chrisjpetito@gmail.com>

Author: krissetto

docs/USAGE.md

@@ -170,6 +170,7 @@ During TUI sessions, you can use special slash commands. Type `/` to see all ava
 | `/sessions` | Browse and load past sessions                                       |
 | `/shell`    | Start a shell                                                       |
 | `/star`     | Toggle star on current session                                      |
+| `/think`    | Toggle thinking/reasoning mode                                      |
 | `/yolo`     | Toggle automatic approval of tool calls                             |
 
 #### Runtime Model Switching

pkg/api/types.go

@@ -131,6 +131,7 @@ type SessionResponse struct {
 	Messages      []session.Message          `json:"messages,omitempty"`
 	CreatedAt     time.Time                  `json:"created_at"`
 	ToolsApproved bool                       `json:"tools_approved"`
+	Thinking      bool                       `json:"thinking"`
 	InputTokens   int64                      `json:"input_tokens"`
 	OutputTokens  int64                      `json:"output_tokens"`
 	WorkingDir    string                     `json:"working_dir,omitempty"`

pkg/model/provider/anthropic/beta_client.go

@@ -70,21 +70,27 @@ func (c *Client) createBetaStream(
 		params.OutputFormat = anthropic.BetaJSONSchemaOutputFormat(structuredOutput.Schema)
 	}
 
+	// Configure thinking if not explicitly disabled via /think command
 	// For interleaved thinking to make sense, we use a default of 16384 tokens for the thinking budget
-	thinkingTokens := int64(16384)
-	if c.ModelConfig.ThinkingBudget != nil {
-		thinkingTokens = int64(c.ModelConfig.ThinkingBudget.Tokens)
+	thinkingEnabled := c.ModelOptions.Thinking() == nil || *c.ModelOptions.Thinking()
+	if thinkingEnabled {
+		thinkingTokens := int64(16384)
+		if c.ModelConfig.ThinkingBudget != nil {
+			thinkingTokens = int64(c.ModelConfig.ThinkingBudget.Tokens)
+		} else {
+			slog.Info("Anthropic Beta API using default thinking_budget with interleaved thinking", "budget_tokens", thinkingTokens)
+		}
+		switch {
+		case thinkingTokens >= 1024 && thinkingTokens < maxTokens:
+			params.Thinking = anthropic.BetaThinkingConfigParamOfEnabled(thinkingTokens)
+			slog.Debug("Anthropic Beta API using thinking_budget with interleaved thinking", "budget_tokens", thinkingTokens)
+		case thinkingTokens >= maxTokens:
+			slog.Warn("Anthropic Beta API thinking_budget must be less than max_tokens, ignoring", "tokens", thinkingTokens, "max_tokens", maxTokens)
+		default:
+			slog.Warn("Anthropic Beta API thinking_budget below minimum (1024), ignoring", "tokens", thinkingTokens)
+		}
 	} else {
-		slog.Info("Anthropic Beta API using default thinking_budget with interleaved thinking", "budget_tokens", thinkingTokens)
-	}
-	switch {
-	case thinkingTokens >= 1024 && thinkingTokens < maxTokens:
-		params.Thinking = anthropic.BetaThinkingConfigParamOfEnabled(thinkingTokens)
-		slog.Debug("Anthropic Beta API using thinking_budget with interleaved thinking", "budget_tokens", thinkingTokens)
-	case thinkingTokens >= maxTokens:
-		slog.Warn("Anthropic Beta API thinking_budget must be less than max_tokens, ignoring", "tokens", thinkingTokens, "max_tokens", maxTokens)
-	default:
-		slog.Warn("Anthropic Beta API thinking_budget below minimum (1024), ignoring", "tokens", thinkingTokens)
+		slog.Debug("Anthropic Beta API: Thinking disabled via /think command")
 	}
 
 	if len(requestTools) > 0 {

pkg/model/provider/bedrock/client_test.go

@@ -770,7 +770,7 @@ func TestBuildInferenceConfig_DisablesTempTopPWhenThinkingEnabled(t *testing.T)
 	assert.Equal(t, int32(64000), *cfg.MaxTokens)
 }
 
-func TestBuildInferenceConfig_SetsTempTopPWhenThinkingDisabled(t *testing.T) {
+func TestBuildInferenceConfig_SetsTempTopPWhenThinkingNotConfigured(t *testing.T) {
 	t.Parallel()
 
 	temp := 0.7

pkg/model/provider/openai/client.go

@@ -358,7 +358,9 @@ func (c *Client) CreateResponseStream(
 
 	// Configure reasoning for models that support it (o-series, gpt-5)
 	// Request detailed reasoning summary to get thinking traces for reasoning models
-	if isOpenAIReasoningModel(c.ModelConfig.Model) {
+	// Skip reasoning configuration entirely if thinking is explicitly disabled (via /think command)
+	thinkingEnabled := c.ModelOptions.Thinking() == nil || *c.ModelOptions.Thinking()
+	if isOpenAIReasoningModel(c.ModelConfig.Model) && thinkingEnabled {
 		params.Reasoning = shared.ReasoningParam{
 			Summary: shared.ReasoningSummaryDetailed,
 		}