feat(llm): support OpenAI Responses API via per-provider api_type

internal/session/manager.go

@@ -204,10 +204,18 @@ func (m *Manager) CurrentThinking() string {
 	return m.provider.Thinking
 }
 
+func (m *Manager) CurrentAPIType() string {
+	if m.provider.APIType == "" {
+		return config.APITypeChat
+	}
+	return m.provider.APIType
+}
+
 func (m *Manager) Info() string {
 	var b strings.Builder
 	fmt.Fprintf(&b, "provider: %s\n", m.CurrentProvider())
 	fmt.Fprintf(&b, "model: %s\n", m.CurrentModel())
+	fmt.Fprintf(&b, "api: %s\n", m.CurrentAPIType())
 	fmt.Fprintf(&b, "thinking: %s\n", m.CurrentThinking())
 	fmt.Fprintf(&b, "providers: %s\n", strings.Join(m.providerNames(), ", "))
 	fmt.Fprintf(&b, "models: %s\n", strings.Join(m.provider.Models, ", "))
@@ -223,7 +231,7 @@ func (m *Manager) SetProvider(name string) (string, error) {
 	}
 	m.providerName = name
 	m.provider = provider
-	m.syncAgent(llm.NewOpenAICompatibleClient(provider.BaseURL, provider.APIKey, m.httpClient))
+	m.syncAgent(providerClient(provider, m.httpClient))
 	return "Switched provider.\n" + m.summary(), nil
 }
 
@@ -257,7 +265,14 @@ func (m *Manager) SetThinking(level string) (string, error) {
 }
 
 func (m *Manager) summary() string {
-	return fmt.Sprintf("provider: %s\nmodel: %s\nthinking: %s", m.CurrentProvider(), m.CurrentModel(), m.CurrentThinking())
+	return fmt.Sprintf("provider: %s\nmodel: %s\napi: %s\nthinking: %s", m.CurrentProvider(), m.CurrentModel(), m.CurrentAPIType(), m.CurrentThinking())
+}
+
+func providerClient(provider config.ProviderConfig, httpClient *http.Client) llm.Provider {
+	if provider.APIType == config.APITypeResponses {
+		return llm.NewOpenAIResponsesClient(provider.BaseURL, provider.APIKey, httpClient)
+	}
+	return llm.NewOpenAICompatibleClient(provider.BaseURL, provider.APIKey, httpClient)
 }
 
 func (m *Manager) maxContextTokens() int {