feat(api): pass app_id to model plugins for provider-side cost attribution

api/core/app/app_context.py

@@ -0,0 +1,40 @@
+"""Request-scoped app context for model plugin dispatch.
+
+Stores the current app_id in a ContextVar so that PluginModelRuntime
+can include it in the HTTP payload to the plugin daemon, without
+requiring changes to the graphon call chain.
+
+Usage in app runners::
+
+    from core.app.app_context import set_current_app_id
+
+    token = set_current_app_id(app_id)
+    try:
+        invoke_result = model_instance.invoke_llm(...)
+    finally:
+        reset_current_app_id(token)
+
+The value is read by PluginModelRuntime.invoke_llm() to tag the
+request with the originating Dify app.
+"""
+
+from __future__ import annotations
+
+from contextvars import ContextVar, Token
+
+_current_app_id: ContextVar[str | None] = ContextVar("_current_app_id", default=None)
+
+
+def get_current_app_id() -> str | None:
+    """Return the app_id for the current execution context, or None."""
+    return _current_app_id.get()
+
+
+def set_current_app_id(app_id: str | None) -> Token[str | None]:
+    """Set the app_id for the current execution context. Returns a token for reset."""
+    return _current_app_id.set(app_id)
+
+
+def reset_current_app_id(token: Token[str | None]) -> None:
+    """Reset the app_id to its previous value."""
+    _current_app_id.reset(token)

api/core/app/apps/advanced_chat/app_runner.py

@@ -92,9 +92,13 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
 
     @trace_span(WorkflowAppRunnerHandler)
     def run(self):
+        from core.app.app_context import reset_current_app_id, set_current_app_id
+
         app_config = self.application_generate_entity.app_config
         app_config = cast(AdvancedChatAppConfig, app_config)
 
+        _app_id_token = set_current_app_id(app_config.app_id)
+
         system_inputs = build_system_variables(
             query=self.application_generate_entity.query,
             files=self.application_generate_entity.files,
@@ -251,6 +255,8 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
         for event in generator:
             self._handle_event(workflow_entry, event)
 
+        reset_current_app_id(_app_id_token)
+
     def handle_input_moderation(
         self,
         app_record: App,

api/core/app/apps/agent_chat/app_runner.py

@@ -224,6 +224,10 @@ class AgentChatAppRunner(AppRunner):
             model_instance=model_instance,
         )
 
+        from core.app.app_context import reset_current_app_id, set_current_app_id
+
+        _app_id_token = set_current_app_id(app_config.app_id)
+
         invoke_result = runner.run(
             message=message,
             query=query,
@@ -240,3 +244,5 @@ class AgentChatAppRunner(AppRunner):
             user_id=application_generate_entity.user_id,
             tenant_id=app_config.tenant_id,
         )
+
+        reset_current_app_id(_app_id_token)

api/core/app/apps/chat/app_runner.py

@@ -218,12 +218,18 @@ class ChatAppRunner(AppRunner):
 
         db.session.close()
 
-        invoke_result = model_instance.invoke_llm(
-            prompt_messages=prompt_messages,
-            model_parameters=application_generate_entity.model_conf.parameters,
-            stop=stop,
-            stream=application_generate_entity.stream,
-        )
+        from core.app.app_context import reset_current_app_id, set_current_app_id
+
+        _app_id_token = set_current_app_id(app_config.app_id)
+        try:
+            invoke_result = model_instance.invoke_llm(
+                prompt_messages=prompt_messages,
+                model_parameters=application_generate_entity.model_conf.parameters,
+                stop=stop,
+                stream=application_generate_entity.stream,
+            )
+        finally:
+            reset_current_app_id(_app_id_token)
 
         # handle invoke result
         self._handle_invoke_result(

api/core/app/apps/completion/app_runner.py

@@ -176,12 +176,18 @@ class CompletionAppRunner(AppRunner):
 
         db.session.close()
 
-        invoke_result = model_instance.invoke_llm(
-            prompt_messages=prompt_messages,
-            model_parameters=application_generate_entity.model_conf.parameters,
-            stop=stop,
-            stream=application_generate_entity.stream,
-        )
+        from core.app.app_context import reset_current_app_id, set_current_app_id
+
+        _app_id_token = set_current_app_id(app_config.app_id)
+        try:
+            invoke_result = model_instance.invoke_llm(
+                prompt_messages=prompt_messages,
+                model_parameters=application_generate_entity.model_conf.parameters,
+                stop=stop,
+                stream=application_generate_entity.stream,
+            )
+        finally:
+            reset_current_app_id(_app_id_token)
 
         # handle invoke result
         self._handle_invoke_result(

api/core/app/apps/workflow/app_runner.py

@@ -64,8 +64,12 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
         """
         Run application
         """
+        from core.app.app_context import reset_current_app_id, set_current_app_id
+
         app_config = self.application_generate_entity.app_config
         app_config = cast(WorkflowAppConfig, app_config)
+
+        _app_id_token = set_current_app_id(app_config.app_id)
         invoke_from = self.application_generate_entity.invoke_from
         # if only single iteration or single loop run is requested
         if self.application_generate_entity.single_iteration_run or self.application_generate_entity.single_loop_run:
@@ -173,3 +177,5 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
 
         for event in generator:
             self._handle_event(workflow_entry, event)
+
+        reset_current_app_id(_app_id_token)

api/core/plugin/impl/model.py

@@ -23,10 +23,14 @@ from graphon.model_runtime.utils.encoders import jsonable_encoder
 
 class PluginModelClient(BasePluginClient):
     @staticmethod
-    def _dispatch_payload(*, user_id: str | None, data: dict[str, Any]) -> dict[str, Any]:
+    def _dispatch_payload(
+        *, user_id: str | None, data: dict[str, Any], app_id: str | None = None
+    ) -> dict[str, Any]:
         payload: dict[str, Any] = {"data": data}
         if user_id is not None:
             payload["user_id"] = user_id
+        if app_id is not None:
+            payload["app_id"] = app_id
         return payload
 
     def fetch_model_providers(self, tenant_id: str) -> Sequence[PluginModelProviderEntity]:
@@ -162,6 +166,7 @@ class PluginModelClient(BasePluginClient):
         tools: list[PromptMessageTool] | None = None,
         stop: list[str] | None = None,
         stream: bool = True,
+        app_id: str | None = None,
     ) -> Generator[LLMResultChunk, None, None]:
         """
         Invoke llm
@@ -173,6 +178,7 @@ class PluginModelClient(BasePluginClient):
             data=jsonable_encoder(
                 self._dispatch_payload(
                     user_id=user_id,
+                    app_id=app_id,
                     data={
                         "provider": provider,
                         "model_type": "llm",

api/core/plugin/impl/model_runtime.py

@@ -207,7 +207,10 @@ class PluginModelRuntime(ModelRuntime):
         stop: Sequence[str] | None,
         stream: bool,
     ) -> Union[LLMResult, Generator[LLMResultChunk, None, None]]:
+        from core.app.app_context import get_current_app_id
+
         plugin_id, provider_name = self._split_provider(provider)
+        app_id = get_current_app_id()
         return self.client.invoke_llm(
             tenant_id=self.tenant_id,
             user_id=self.user_id,
@@ -220,6 +223,7 @@ class PluginModelRuntime(ModelRuntime):
             tools=tools,
             stop=list(stop) if stop else None,
             stream=stream,
+            app_id=app_id,
         )
 
     def get_llm_num_tokens(

api/tests/unit_tests/core/plugin/impl/test_model_client.py

@@ -160,6 +160,58 @@ class TestPluginModelClient:
         assert call_kwargs["data"]["data"]["stream"] is False
         assert call_kwargs["data"]["data"]["model_parameters"] == {"temperature": 0.1}
 
+    def test_invoke_llm_with_app_id(self, mocker):
+        client = PluginModelClient()
+        stream_mock = mocker.patch.object(
+            client, "_request_with_plugin_daemon_response_stream", return_value=iter(["chunk-1"])
+        )
+
+        list(
+            client.invoke_llm(
+                tenant_id="tenant-1",
+                user_id="user-1",
+                plugin_id="org/plugin:1",
+                provider="provider-a",
+                model="gpt-test",
+                credentials={"api_key": "key"},
+                prompt_messages=[],
+                app_id="app-123",
+            )
+        )
+
+        call_kwargs = stream_mock.call_args.kwargs
+        assert call_kwargs["data"]["app_id"] == "app-123"
+
+    def test_invoke_llm_without_app_id_omits_field(self, mocker):
+        client = PluginModelClient()
+        stream_mock = mocker.patch.object(
+            client, "_request_with_plugin_daemon_response_stream", return_value=iter(["chunk-1"])
+        )
+
+        list(
+            client.invoke_llm(
+                tenant_id="tenant-1",
+                user_id="user-1",
+                plugin_id="org/plugin:1",
+                provider="provider-a",
+                model="gpt-test",
+                credentials={"api_key": "key"},
+                prompt_messages=[],
+            )
+        )
+
+        call_kwargs = stream_mock.call_args.kwargs
+        assert "app_id" not in call_kwargs["data"]
+
+    def test_dispatch_payload_includes_app_id_when_provided(self):
+        payload = PluginModelClient._dispatch_payload(user_id="u1", data={"k": "v"}, app_id="app-456")
+        assert payload["app_id"] == "app-456"
+        assert payload["user_id"] == "u1"
+
+    def test_dispatch_payload_omits_app_id_when_none(self):
+        payload = PluginModelClient._dispatch_payload(user_id="u1", data={"k": "v"})
+        assert "app_id" not in payload
+
     def test_invoke_llm_wraps_plugin_daemon_inner_error(self, mocker):
         client = PluginModelClient()