fix: add no_think option to disable qwen3 thinking mode for classifier

qwen3 returns empty content when thinking mode is active, causing
classifier to fail. no_think: true sends think=false in Ollama options.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Alexey Panfilov

config/config.yaml

@@ -74,6 +74,7 @@ models:
     model: qwen3:0.6b
     base_url: http://host.docker.internal:11434
     max_tokens: 64
+    no_think: true
 
   # Ollama local — simple tasks via Ollama Cloud model (routed through local Ollama).
   ollama-local:

internal/config/config.go

@@ -69,7 +69,8 @@ type ModelConfig struct {
 	APIKey    string `yaml:"api_key"`
 	MaxTokens int    `yaml:"max_tokens"`
 	BaseURL   string `yaml:"base_url"`
-	Vision    bool   `yaml:"vision"` // true if the model supports image input
+	Vision    bool   `yaml:"vision"`    // true if the model supports image input
+	NoThink   bool   `yaml:"no_think"`  // disable thinking mode (qwen3)
 }
 
 type ModelsConfig struct {

internal/llm/ollama.go

@@ -23,6 +23,7 @@ type ollamaProvider struct {
 	maxTokens int
 	provName  string
 	vision    bool
+	noThink   bool
 }
 
 // NewOllama creates a provider for a local Ollama instance.
@@ -43,6 +44,7 @@ func NewOllama(cfg config.ModelConfig) (*ollamaProvider, error) {
 		maxTokens: maxTokens,
 		provName:  "ollama",
 		vision:    cfg.Vision,
+		noThink:   cfg.NoThink,
 	}, nil
 }
 
@@ -92,7 +94,8 @@ type ollamaFunctionDef struct {
 }
 
 type ollamaOptions struct {
-	NumPredict int `json:"num_predict,omitempty"`
+	NumPredict int   `json:"num_predict,omitempty"`
+	Think      *bool `json:"think,omitempty"`
 }
 
 type ollamaResponse struct {
@@ -105,11 +108,16 @@ type ollamaResponse struct {
 func (p *ollamaProvider) Chat(ctx context.Context, messages []Message, systemPrompt string, tools []Tool) (Response, error) {
 	ollamaMsgs := p.buildMessages(messages, systemPrompt)
 
+	opts := &ollamaOptions{NumPredict: p.maxTokens}
+	if p.noThink {
+		f := false
+		opts.Think = &f
+	}
 	req := ollamaRequest{
 		Model:    p.model,
 		Messages: ollamaMsgs,
 		Stream:   false,
-		Options:  &ollamaOptions{NumPredict: p.maxTokens},
+		Options:  opts,
 	}
 	if len(tools) > 0 {
 		req.Tools = buildOllamaTools(tools)