fix(#3996): preserve Gemini user thinking budgets outside title generation

pkg/model/provider/gemini/client.go

@@ -425,7 +425,7 @@ func extractMimeType(dataURLPrefix string) string {
 	return "image/jpeg" // Default fallback
 }
 
-// buildConfig creates GenerateContentConfig from model config
+// BuildConfig creates GenerateContentConfig from model config.
 func (c *Client) buildConfig() *genai.GenerateContentConfig {
 	config := &genai.GenerateContentConfig{}
 	if c.ModelConfig.MaxTokens != nil {
@@ -453,7 +453,11 @@ func (c *Client) buildConfig() *genai.GenerateContentConfig {
 	// Apply thinking configuration for Gemini models.
 	// See https://ai.google.dev/gemini-api/docs/thinking
 	if c.ModelOptions.NoThinking() {
-		// NoThinking requested (e.g. title generation). For Gemini 3+ models
+		if c.ModelOptions.GeneratingTitle() {
+			return config
+		}
+
+		// NoThinking requested (e.g. MCP sampling). For Gemini 3+ models
 		// that always think, use the lowest level and bump MaxOutputTokens so
 		// internal reasoning doesn't consume the entire budget. Gemini 2.5 and
 		// older can fully disable thinking with ThinkingBudget=0.

pkg/model/provider/gemini/client_test.go

@@ -10,10 +10,74 @@ import (
 	"github.com/docker/docker-agent/pkg/chat"
 	"github.com/docker/docker-agent/pkg/config/latest"
 	"github.com/docker/docker-agent/pkg/model/provider/base"
+	"github.com/docker/docker-agent/pkg/model/provider/options"
 	"github.com/docker/docker-agent/pkg/modelsdev"
 	"github.com/docker/docker-agent/pkg/tools"
 )
 
+func TestBuildConfig_NoThinking(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name          string
+		model         string
+		opts          []options.Opt
+		wantThinking  bool
+		wantMinTokens bool
+	}{
+		{
+			name:         "title generation omits thinking config",
+			model:        "gemini-3-flash",
+			opts:         []options.Opt{options.WithGeneratingTitle(), options.WithNoThinking()},
+			wantThinking: false,
+		},
+		{
+			name:          "MCP sampling disables Gemini 3 thinking",
+			model:         "gemini-3-flash",
+			opts:          []options.Opt{options.WithNoThinking()},
+			wantThinking:  true,
+			wantMinTokens: true,
+		},
+		{
+			name:         "MCP sampling disables Gemini 2.5 thinking",
+			model:        "gemini-2.5-flash",
+			opts:         []options.Opt{options.WithNoThinking()},
+			wantThinking: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			client := &Client{Config: base.Config{
+				ModelConfig: latest.ModelConfig{
+					Provider:       "google",
+					Model:          tt.model,
+					ThinkingBudget: &latest.ThinkingBudget{Effort: "high"},
+				},
+				ModelOptions: options.Apply(tt.opts...),
+			}}
+
+			config := client.buildConfig()
+			if !tt.wantThinking {
+				assert.Nil(t, config.ThinkingConfig)
+				return
+			}
+
+			require.NotNil(t, config.ThinkingConfig)
+			assert.False(t, config.ThinkingConfig.IncludeThoughts)
+			if tt.wantMinTokens {
+				assert.Equal(t, genai.ThinkingLevelLow, config.ThinkingConfig.ThinkingLevel)
+				assert.GreaterOrEqual(t, config.MaxOutputTokens, int32(200))
+				return
+			}
+			require.NotNil(t, config.ThinkingConfig.ThinkingBudget)
+			assert.Zero(t, *config.ThinkingConfig.ThinkingBudget)
+		})
+	}
+}
+
 func TestBuildConfig_Gemini25_ThinkingBudget(t *testing.T) {
 	t.Parallel()