Add fastgpt as seperate assistant mode

engine/runtime/ports/__init__.py

@@ -14,7 +14,13 @@ from runtime.ports.control_plane import (
     KnowledgeRetriever,
     ToolCatalog,
 )
-from runtime.ports.llm import LLMCancellable, LLMPort, LLMRuntimeConfigurable, LLMServiceSpec
+from runtime.ports.llm import (
+    LLMCancellable,
+    LLMClientToolResumable,
+    LLMPort,
+    LLMRuntimeConfigurable,
+    LLMServiceSpec,
+)
 from runtime.ports.service_factory import RealtimeServiceFactory
 from runtime.ports.tts import TTSPort, TTSServiceSpec
 
@@ -30,6 +36,7 @@ __all__ = [
     "KnowledgeRetriever",
     "ToolCatalog",
     "LLMCancellable",
+    "LLMClientToolResumable",
     "LLMPort",
     "LLMRuntimeConfigurable",
     "LLMServiceSpec",

engine/runtime/ports/llm.py

@@ -18,6 +18,7 @@ class LLMServiceSpec:
     model: str
     api_key: Optional[str] = None
     base_url: Optional[str] = None
+    app_id: Optional[str] = None
     system_prompt: Optional[str] = None
     temperature: float = 0.7
     knowledge_config: Dict[str, Any] = field(default_factory=dict)
@@ -65,3 +66,17 @@ class LLMRuntimeConfigurable(Protocol):
 
     def set_tool_schemas(self, schemas: Optional[List[Dict[str, Any]]]) -> None:
         """Apply runtime tool schemas used for tool calling."""
+
+
+class LLMClientToolResumable(Protocol):
+    """Optional extension for providers that pause on client-side tool results."""
+
+    def handles_client_tool(self, tool_name: str) -> bool:
+        """Return True when the provider owns the lifecycle of this client tool."""
+
+    def resume_after_client_tool_result(
+        self,
+        tool_call_id: str,
+        result: Dict[str, Any],
+    ) -> AsyncIterator[LLMStreamEvent]:
+        """Resume the provider stream after a correlated client-side tool result."""