Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

fix(agent): meter internal LLM calls through API (#40613)

zyssyz123c9c7fb9962d3
diff --git a/api/clients/agent_backend/request_builder.py b/api/clients/agent_backend/request_builder.py
index 61d74eb4a4b..984aeef2174 100644
--- a/api/clients/agent_backend/request_builder.py
+++ b/api/clients/agent_backend/request_builder.py
@@ -25,7 +25,6 @@ from dify_agent.layers.dify_core_tools import DIFY_CORE_TOOLS_LAYER_TYPE_ID, Dif
from dify_agent.layers.dify_plugin import (
DIFY_PLUGIN_LLM_LAYER_TYPE_ID,
DIFY_PLUGIN_TOOLS_LAYER_TYPE_ID,
- DifyPluginCredentialValue,
DifyPluginLLMLayerConfig,
DifyPluginToolsLayerConfig,
)
@@ -150,7 +149,6 @@ class AgentBackendModelConfig(BaseModel):
plugin_id: str
model_provider: str
model: str
- credentials: dict[str, DifyPluginCredentialValue] = Field(default_factory=dict)
model_settings: dict[str, JsonValue] = Field(default_factory=dict)
model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid")
@@ -414,7 +412,6 @@ class AgentBackendRunRequestBuilder:
plugin_id=run_input.model.plugin_id,
model_provider=run_input.model.model_provider,
model=run_input.model.model,
- credentials=run_input.model.credentials,
model_settings=_agent_model_settings(run_input.model.model_settings),
),
)
@@ -607,7 +604,6 @@ class AgentBackendRunRequestBuilder:
plugin_id=run_input.model.plugin_id,
model_provider=run_input.model.model_provider,
model=run_input.model.model,
- credentials=run_input.model.credentials,
model_settings=_agent_model_settings(run_input.model.model_settings),
),
),
diff --git a/api/controllers/inner_api/__init__.py b/api/controllers/inner_api/__init__.py
index 277e0f5ec0e..32dabf5bb01 100644
--- a/api/controllers/inner_api/__init__.py
+++ b/api/controllers/inner_api/__init__.py
@@ -18,6 +18,7 @@ inner_api_ns = Namespace("inner_api", description="Internal API operations", pat
from . import mail as _mail
from . import runtime_credentials as _runtime_credentials
from .agent import files as _agent_files
+from .agent import llm as _agent_llm
from .agent import tools as _agent_tools
from .app import dsl as _app_dsl
from .knowledge import retrieval as _knowledge_retrieval
@@ -32,6 +33,7 @@ __all__ = [
"_agent_config",
"_agent_drive",
"_agent_files",
+ "_agent_llm",
"_agent_tools",
"_app_dsl",
"_knowledge_retrieval",
diff --git a/api/controllers/inner_api/agent/llm.py b/api/controllers/inner_api/agent/llm.py
new file mode 100644
index 00000000000..c4eaa64358a
--- /dev/null
+++ b/api/controllers/inner_api/agent/llm.py
@@ -0,0 +1,113 @@
+"""Trusted streaming LLM gateway for dify-agent runs."""
+
+from __future__ import annotations
+
+import json
+from collections.abc import Generator
+
+from flask import Response, stream_with_context
+from flask_restx import Resource
+from pydantic import ValidationError
+
+from controllers.common.schema import register_schema_models
+from controllers.inner_api import inner_api_ns
+from controllers.inner_api.wraps import agent_inner_api_only
+from core.errors.error import QuotaExceededError
+from libs.exception import BaseHTTPException
+from services.agent_llm_inner_service import AgentLLMInnerService, AgentLLMInnerServiceError
+from services.entities.agent_llm_inner import AgentLLMInvokeRequest
+
+
+class AgentLLMInvokeHttpError(BaseHTTPException):
+ error_code = "agent_llm_invoke_failed"
+ description = "Agent LLM invocation failed."
+ code = 500
+
+ def __init__(self, *, error_code: str, description: str, status_code: int) -> None:
+ self.error_code = error_code
+ self.description = description
+ self.code = status_code
+ super().__init__(description)
+
+
+register_schema_models(inner_api_ns, AgentLLMInvokeRequest)
+
+
+@inner_api_ns.route("/agent/llm/invoke")
+class AgentLLMInvokeApi(Resource):
+ """Resolve one dify-agent model request before proxying it."""
+
+ @agent_inner_api_only
+ @inner_api_ns.doc("inner_agent_llm_invoke")
+ @inner_api_ns.expect(inner_api_ns.models[AgentLLMInvokeRequest.__name__])
+ @inner_api_ns.produces(["text/event-stream"])
+ def post(self) -> Response:
+ try:
+ payload = AgentLLMInvokeRequest.model_validate(inner_api_ns.payload or {})
+ except ValidationError as exc:
+ raise AgentLLMInvokeHttpError(
+ error_code="invalid_request",
+ description=str(exc),
+ status_code=400,
+ ) from exc
+
+ service = AgentLLMInnerService()
+ try:
+ prepared = service.prepare(payload)
+ except AgentLLMInnerServiceError as exc:
+ raise AgentLLMInvokeHttpError(
+ error_code=exc.error_code,
+ description=exc.description,
+ status_code=exc.status_code,
+ ) from exc
+ except QuotaExceededError as exc:
+ raise AgentLLMInvokeHttpError(
+ error_code="agent_llm_quota_exceeded",
+ description=str(exc) or "Insufficient Message Credits.",
+ status_code=429,
+ ) from exc
+ except ValueError as exc:
+ raise AgentLLMInvokeHttpError(
+ error_code="invalid_model_request",
+ description=str(exc),
+ status_code=400,
+ ) from exc
+
+ def generate() -> Generator[str, None, None]:
+ try:
+ for chunk in service.invoke(prepared):
+ envelope = {"code": 0, "message": "", "data": chunk.model_dump(mode="json")}
+ yield f"data: {json.dumps(envelope, ensure_ascii=False, separators=(',', ':'))}\n\n"
+ except GeneratorExit:
+ raise
+ except QuotaExceededError as exc:
+ error = {
+ "error_type": "AgentLLMQuotaExceededError",
+ "message": str(exc) or "Insufficient hosted model quota remaining.",
+ }
+ envelope = {
+ "code": -429,
+ "message": json.dumps(error, ensure_ascii=False, separators=(",", ":")),
+ "data": None,
+ }
+ yield f"data: {json.dumps(envelope, ensure_ascii=False, separators=(',', ':'))}\n\n"
+ except Exception as exc:
+ error = {
+ "error_type": type(exc).__name__,
+ "message": str(exc) or "Agent LLM invocation failed.",
+ }
+ envelope = {
+ "code": -500,
+ "message": json.dumps(error, ensure_ascii=False, separators=(",", ":")),
+ "data": None,
+ }
+ yield f"data: {json.dumps(envelope, ensure_ascii=False, separators=(',', ':'))}\n\n"
+
+ return Response(
+ stream_with_context(generate()), # pyrefly: ignore[no-matching-overload]
+ content_type="text/event-stream",
+ headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
+ )
+
+
+__all__ = ["AgentLLMInvokeApi", "AgentLLMInvokeHttpError"]
diff --git a/api/core/app/apps/agent_app/app_generator.py b/api/core/app/apps/agent_app/app_generator.py
index a291b84b39b..6752690b0d3 100644
--- a/api/core/app/apps/agent_app/app_generator.py
+++ b/api/core/app/apps/agent_app/app_generator.py
@@ -45,7 +45,6 @@ from core.app.entities.app_invoke_entities import (
InvokeFrom,
UserFrom,
)
-from core.app.llm.model_access import build_dify_model_access
from core.db.session_factory import session_factory
from core.ops.ops_trace_manager import TraceQueueManager
from core.workflow.file_reference import build_file_reference, is_canonical_file_reference
@@ -219,6 +218,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
agent_config_snapshot_id=agent_config_id,
agent_config_version_kind=agent_config_version_kind,
agent_session_scope_config_version_id=session_scope_config_version_id,
+ agent_llm_gateway_enabled=True,
)
conversation, message = self._init_generate_records(
@@ -349,6 +349,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
agent_id=agent.id,
agent_config_snapshot_id=agent_config_id,
agent_config_version_kind=agent_config_version_kind,
+ agent_llm_gateway_enabled=True,
)
conversation, message = self._init_generate_records(
@@ -506,7 +507,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
session=session,
)
- runner = self._build_runner(dify_context)
+ runner = self._build_runner()
runner.run(
dify_context=dify_context,
agent_id=application_generate_entity.agent_id,
@@ -542,10 +543,9 @@ class AgentAppGenerator(MessageBasedAppGenerator):
return query.replace("\x00", "")
@staticmethod
- def _build_runner(dify_context: DifyRunContext) -> AgentAppRunner:
- credentials_provider, _ = build_dify_model_access(dify_context)
+ def _build_runner() -> AgentAppRunner:
return AgentAppRunner(
- request_builder=AgentAppRuntimeRequestBuilder(credentials_provider=credentials_provider),
+ request_builder=AgentAppRuntimeRequestBuilder(),
agent_backend_client=create_agent_backend_run_client(
base_url=dify_config.AGENT_BACKEND_BASE_URL,
api_token=dify_config.AGENT_BACKEND_API_TOKEN,
diff --git a/api/core/app/apps/agent_app/runtime_request_builder.py b/api/core/app/apps/agent_app/runtime_request_builder.py
index bab7a0059bb..6fc85435c5d 100644
--- a/api/core/app/apps/agent_app/runtime_request_builder.py
+++ b/api/core/app/apps/agent_app/runtime_request_builder.py
@@ -10,9 +10,8 @@ used by workflow runs.
from __future__ import annotations
-from collections.abc import Mapping
from dataclasses import dataclass
-from typing import Any, Literal, Protocol, cast
+from typing import Any, Literal, cast
from agenton.compositor import CompositorSessionSnapshot
from dify_agent.layers.execution_context import (
@@ -58,10 +57,6 @@ class AgentAppRuntimeRequestBuildError(ValueError):
super().__init__(message)
-class CredentialsProvider(Protocol):
- def fetch(self, provider_name: str, model_name: str) -> dict[str, Any]: ...
-
-
@dataclass(frozen=True, slots=True)
class AgentAppRuntimeBuildContext:
dify_context: DifyRunContext
@@ -93,11 +88,9 @@ class AgentAppRuntimeRequestBuilder:
def __init__(
self,
*,
- credentials_provider: CredentialsProvider,
request_builder: AgentBackendRunRequestBuilder | None = None,
dify_tools_builder: WorkflowAgentDifyToolLayersBuilder | None = None,
) -> None:
- self._credentials_provider = credentials_provider
self._request_builder = request_builder or AgentBackendRunRequestBuilder()
self._dify_tools_builder = dify_tools_builder or WorkflowAgentDifyToolsBuilder()
@@ -110,7 +103,6 @@ class AgentAppRuntimeRequestBuilder:
)
metadata = self._build_metadata(context)
- credentials = self._credentials_provider.fetch(agent_soul.model.model_provider, agent_soul.model.model)
try:
tool_layers = self._build_tool_layers(
tenant_id=context.dify_context.tenant_id,
@@ -148,7 +140,6 @@ class AgentAppRuntimeRequestBuilder:
plugin_id=model_plugin_id,
model_provider=model_provider,
model=agent_soul.model.model,
- credentials=self._normalize_credentials(credentials),
model_settings=agent_soul.model.model_settings.model_dump(mode="json", exclude_none=True),
),
execution_context=DifyExecutionContextLayerConfig(
@@ -222,16 +213,6 @@ class AgentAppRuntimeRequestBuilder:
"agent_config_snapshot_id": context.agent_config_snapshot_id,
}
- @staticmethod
- def _normalize_credentials(credentials: Mapping[str, Any]) -> dict[str, str | int | float | bool | None]:
- normalized: dict[str, str | int | float | bool | None] = {}
- for key, value in credentials.items():
- if isinstance(value, str | int | float | bool) or value is None:
- normalized[key] = value
- else:
- normalized[key] = str(value)
- return normalized
-
__all__ = [
"AgentAppRuntimeBuildContext",
diff --git a/api/core/app/entities/app_invoke_entities.py b/api/core/app/entities/app_invoke_entities.py
index b4e2ced08aa..c59208c4aaf 100644
--- a/api/core/app/entities/app_invoke_entities.py
+++ b/api/core/app/entities/app_invoke_entities.py
@@ -238,6 +238,7 @@ class AgentAppGenerateEntity(ChatAppGenerateEntity):
agent_config_version_kind: Literal["snapshot", "draft", "build_draft"] = "snapshot"
agent_session_scope_config_version_id: str | None = None
prompt_file_mappings: Sequence[JsonValue] = Field(default_factory=list)
+ agent_llm_gateway_enabled: bool = False
class AdvancedChatAppGenerateEntity(ConversationAppGenerateEntity):
diff --git a/api/core/app/llm/quota.py b/api/core/app/llm/quota.py
index ad28f13a6af..2c892b9fd98 100644
--- a/api/core/app/llm/quota.py
+++ b/api/core/app/llm/quota.py
@@ -115,7 +115,12 @@ def _get_current_quota_configuration(system_configuration):
def reserve_model_quota_for_model(
- *, tenant_id: str, provider: str, model_type: ModelType, model: str
+ *,
+ tenant_id: str,
+ provider: str,
+ model_type: ModelType,
+ model: str,
+ request_id: str | None = None,
) -> ModelQuotaReservation:
"""Reserve system-hosted model quota before invoking the provider."""
provider_configuration = _get_provider_configuration(tenant_id=tenant_id, provider=provider)
@@ -165,7 +170,7 @@ def reserve_model_quota_for_model(
tenant_id=tenant_id,
credits_required=amount,
pool_type="paid" if quota_type == ProviderQuotaType.PAID else "trial",
- request_id=str(uuid4()),
+ request_id=request_id or str(uuid4()),
session_factory=db.session,
meta=reservation_meta,
)
@@ -177,13 +182,16 @@ def reserve_model_quota_for_model(
return reservation
-def reserve_llm_quota_for_model(*, tenant_id: str, provider: str, model: str) -> ModelQuotaReservation:
+def reserve_llm_quota_for_model(
+ *, tenant_id: str, provider: str, model: str, request_id: str | None = None
+) -> ModelQuotaReservation:
"""Reserve system-hosted LLM quota before invoking the provider."""
return reserve_model_quota_for_model(
tenant_id=tenant_id,
provider=provider,
model_type=ModelType.LLM,
model=model,
+ request_id=request_id,
)
diff --git a/api/core/model_manager.py b/api/core/model_manager.py
index 00059d2a35b..2e91e4fe507 100644
--- a/api/core/model_manager.py
+++ b/api/core/model_manager.py
@@ -2,6 +2,7 @@ import logging
from collections.abc import Callable, Generator, Iterable, Mapping, Sequence
from copy import deepcopy
from typing import IO, Any, Literal, Optional, ParamSpec, TypeVar, Union, cast, overload, override
+from uuid import UUID
from configs import dify_config
from core.entities import PluginCredentialType
@@ -445,7 +446,7 @@ class ModelInstance:
class QuotaManagedModelInstance(ModelInstance):
"""A system-hosted model instance that owns quota settlement per invocation."""
- def reserve_quota(self):
+ def reserve_quota(self, *, request_id: str | None = None):
from core.app.llm.quota import reserve_model_quota_for_model
return reserve_model_quota_for_model(
@@ -453,8 +454,25 @@ class QuotaManagedModelInstance(ModelInstance):
provider=self.provider,
model_type=self.model_type_instance.model_type,
model=self.model_name,
+ request_id=request_id,
)
+ @staticmethod
+ def _get_reservation_request_id(request_metadata: Mapping[str, object] | None) -> str | None:
+ request_id = request_metadata.get("invocation_id") if request_metadata else None
+ if not isinstance(request_id, str) or not request_id:
+ return None
+ try:
+ return str(UUID(request_id))
+ except ValueError:
+ return None
+
+ def _reserve_quota_for_request(self, request_metadata: Mapping[str, object] | None):
+ request_id = self._get_reservation_request_id(request_metadata)
+ if request_id is None:
+ return self.reserve_quota()
+ return self.reserve_quota(request_id=request_id)
+
@staticmethod
def release_quota_safely(reservation) -> None:
try:
@@ -530,7 +548,7 @@ class QuotaManagedModelInstance(ModelInstance):
request_metadata=request_metadata,
)
- reservation = self.reserve_quota()
+ reservation = self._reserve_quota_for_request(request_metadata)
try:
response = super().invoke_llm(
prompt_messages=normalized_prompt_messages,
@@ -558,7 +576,7 @@ class QuotaManagedModelInstance(ModelInstance):
callbacks: list[Callback] | None,
request_metadata: Mapping[str, object] | None,
) -> Generator:
- reservation = self.reserve_quota()
+ reservation = self._reserve_quota_for_request(request_metadata)
usage: LLMUsage | None = None
try:
response = super().invoke_llm(
diff --git a/api/core/workflow/node_factory.py b/api/core/workflow/node_factory.py
index f73dd47e7bc..5d1a20711a7 100644
--- a/api/core/workflow/node_factory.py
+++ b/api/core/workflow/node_factory.py
@@ -530,7 +530,6 @@ class DifyNodeFactory(NodeFactory):
return {
"binding_resolver": WorkflowAgentBindingResolver(),
"runtime_request_builder": WorkflowAgentRuntimeRequestBuilder(
- credentials_provider=self._llm_credentials_provider,
request_builder=AgentBackendRunRequestBuilder(),
),
"agent_backend_client": create_agent_backend_run_client(
diff --git a/api/core/workflow/nodes/agent_v2/runtime_request_builder.py b/api/core/workflow/nodes/agent_v2/runtime_request_builder.py
index 848fd4f80bc..392c04a9125 100644
--- a/api/core/workflow/nodes/agent_v2/runtime_request_builder.py
+++ b/api/core/workflow/nodes/agent_v2/runtime_request_builder.py
@@ -121,10 +121,6 @@ class VariablePoolReader(Protocol):
def get_by_prefix(self, prefix: str, /) -> Mapping[str, object]: ...
-class CredentialsProvider(Protocol):
- def fetch(self, provider_name: str, model_name: str) -> dict[str, Any]: ...
-
-
@dataclass(frozen=True, slots=True)
class WorkflowAgentRuntimeBuildContext:
dify_context: DifyRunContext
@@ -165,11 +161,9 @@ class WorkflowAgentRuntimeRequestBuilder:
def __init__(
self,
*,
- credentials_provider: CredentialsProvider,
request_builder: AgentBackendRunRequestBuilder | None = None,
dify_tools_builder: WorkflowAgentDifyToolLayersBuilder | None = None,
) -> None:
- self._credentials_provider = credentials_provider
self._request_builder = request_builder or AgentBackendRunRequestBuilder()
self._dify_tools_builder = dify_tools_builder or WorkflowAgentDifyToolsBuilder()
@@ -190,7 +184,6 @@ class WorkflowAgentRuntimeRequestBuilder:
workflow_context_prompt = self._build_workflow_context_prompt(context, effective_node_job)
workflow_job_prompt = workflow_task_prompt or self._WORKFLOW_JOB_PROMPT_FALLBACK
user_prompt = workflow_context_prompt or self._WORKFLOW_USER_PROMPT_FALLBACK
- credentials = self._credentials_provider.fetch(agent_soul.model.model_provider, agent_soul.model.model)
try:
tool_layers = self._build_tool_layers(
tenant_id=context.dify_context.tenant_id,
@@ -229,15 +222,14 @@ class WorkflowAgentRuntimeRequestBuilder:
plugin_id=model_plugin_id,
model_provider=model_provider,
model=agent_soul.model.model,
- credentials=self._normalize_credentials(credentials),
model_settings=agent_soul.model.model_settings.model_dump(mode="json", exclude_none=True),
),
# The execution-context layer is now the only public protocol
# carrier for Dify tenant/user/run identifiers. ``user_id`` and
# ``user_from`` must be forwarded here because downstream plugin-
- # daemon provider/tool clients and knowledge-base layers read
- # caller identity from this layer rather than from any parallel
- # top-level request field.
+ # API model gateway, daemon tool clients, and knowledge-base
+ # layers read caller identity from this layer rather than from
+ # any parallel top-level request field.
execution_context=DifyExecutionContextLayerConfig(
tenant_id=context.dify_context.tenant_id,
user_id=context.dify_context.user_id,
@@ -716,16 +708,6 @@ class WorkflowAgentRuntimeRequestBuilder:
if required:
schema["required"] = required
- @staticmethod
- def _normalize_credentials(credentials: Mapping[str, Any]) -> dict[str, str | int | float | bool | None]:
- normalized: dict[str, str | int | float | bool | None] = {}
- for key, value in credentials.items():
- if isinstance(value, str | int | float | bool) or value is None:
- normalized[key] = value
- else:
- normalized[key] = str(value)
- return normalized
-
def build_shell_layer_config(agent_soul: AgentSoulConfig) -> DifyShellLayerConfig:
"""Map Agent Soul shell-adjacent fields into the Agent backend shell config."""
diff --git a/api/events/event_handlers/update_provider_when_message_created.py b/api/events/event_handlers/update_provider_when_message_created.py
index 15b40afdbf2..a2ed5568c99 100644
--- a/api/events/event_handlers/update_provider_when_message_created.py
+++ b/api/events/event_handlers/update_provider_when_message_created.py
@@ -9,7 +9,11 @@ from sqlalchemy.engine import CursorResult
from sqlalchemy.orm import Session
from configs import dify_config
-from core.app.entities.app_invoke_entities import AgentChatAppGenerateEntity, ChatAppGenerateEntity
+from core.app.entities.app_invoke_entities import (
+ AgentAppGenerateEntity,
+ AgentChatAppGenerateEntity,
+ ChatAppGenerateEntity,
+)
from core.entities.provider_entities import ProviderQuotaType, QuotaUnit, SystemConfiguration
from events.message_event import message_was_created
from extensions.ext_database import db
@@ -121,8 +125,13 @@ def handle(sender: Message, **kwargs):
provider_model_bundle = model_config.provider_model_bundle
provider_configuration = provider_model_bundle.configuration
+ agent_gateway_metered = (
+ isinstance(application_generate_entity, AgentAppGenerateEntity)
+ and application_generate_entity.agent_llm_gateway_enabled
+ )
if (
- provider_configuration.using_provider_type == ProviderType.SYSTEM
+ not agent_gateway_metered
+ and provider_configuration.using_provider_type == ProviderType.SYSTEM
and provider_configuration.system_configuration
and provider_configuration.system_configuration.current_quota_type is not None
):
diff --git a/api/services/agent_llm_inner_service.py b/api/services/agent_llm_inner_service.py
new file mode 100644
index 00000000000..8e4fd80f24e
--- /dev/null
+++ b/api/services/agent_llm_inner_service.py
@@ -0,0 +1,113 @@
+"""API-owned model invocation for dify-agent."""
+
+from __future__ import annotations
+
+from collections.abc import Callable, Generator
+from dataclasses import dataclass
+from typing import cast
+
+from sqlalchemy.orm import Session
+
+from core.db.session_factory import session_factory as default_session_factory
+from core.entities.model_entities import ModelStatus
+from core.model_manager import ModelInstance, ModelManager
+from core.plugin.impl.model_runtime_factory import create_plugin_provider_manager
+from graphon.model_runtime.entities.llm_entities import LLMResultChunk
+from graphon.model_runtime.entities.message_entities import PromptMessage
+from graphon.model_runtime.entities.model_entities import ModelType
+from models.model import App
+from services.entities.agent_llm_inner import AgentLLMInvokeRequest
+
+
+class AgentLLMInnerServiceError(RuntimeError):
+ def __init__(self, error_code: str, description: str, *, status_code: int = 500) -> None:
+ self.error_code = error_code
+ self.description = description
+ self.status_code = status_code
+ super().__init__(description)
+
+
+@dataclass(frozen=True)
+class PreparedAgentLLMInvocation:
+ request: AgentLLMInvokeRequest
+ model_instance: ModelInstance
+
+
+class AgentLLMInnerService:
+ """Resolve the model and invoke it through the API model runtime."""
+
+ def __init__(self, *, session_factory: Callable[[], Session] | None = None) -> None:
+ self._session_factory = session_factory or default_session_factory.create_session
+
+ def prepare(self, request: AgentLLMInvokeRequest) -> PreparedAgentLLMInvocation:
+ caller = request.caller
+ target = request.target
+ self._validate_app_tenant(app_id=caller.app_id, tenant_id=caller.tenant_id)
+ provider_manager = create_plugin_provider_manager(tenant_id=caller.tenant_id, user_id=caller.user_id)
+ model_manager = ModelManager(provider_manager=provider_manager)
+ model_instance = model_manager.get_model_instance(
+ tenant_id=caller.tenant_id,
+ provider=target.provider,
+ model_type=ModelType.LLM,
+ model=target.model,
+ )
+
+ provider_configuration = model_instance.provider_model_bundle.configuration
+ provider_model = provider_configuration.get_provider_model(model_type=ModelType.LLM, model=target.model)
+ if provider_model is None:
+ raise AgentLLMInnerServiceError(
+ "model_not_found",
+ f"Model {target.model} does not exist for provider {target.provider}.",
+ status_code=404,
+ )
+ # The cached model status can lag behind the authoritative reservation.
+ # Preserve other provider errors, but let quota reservation produce 429.
+ if provider_model.status != ModelStatus.QUOTA_EXCEEDED:
+ provider_model.raise_for_status()
+
+ return PreparedAgentLLMInvocation(request=request, model_instance=model_instance)
+
+ def invoke(self, prepared: PreparedAgentLLMInvocation) -> Generator[LLMResultChunk, None, None]:
+ request = prepared.request
+ caller = request.caller
+ target = request.target
+ result = prepared.model_instance.invoke_llm(
+ prompt_messages=cast(list[PromptMessage], target.prompt_messages),
+ model_parameters=target.model_parameters,
+ tools=target.tools,
+ stop=target.stop,
+ # The gateway transport is always streamed, including Pydantic AI's
+ # non-streaming request path, so one response protocol is sufficient.
+ stream=True,
+ request_metadata={
+ "source": "agent_llm_gateway",
+ "invocation_id": caller.invocation_id,
+ "agent_run_id": caller.agent_run_id,
+ "agent_mode": caller.agent_mode,
+ "call_index": caller.call_index,
+ "app_id": caller.app_id,
+ "workflow_run_id": caller.workflow_run_id,
+ "node_execution_id": caller.node_execution_id,
+ "trace_id": caller.trace_id,
+ },
+ )
+ yield from cast(Generator[LLMResultChunk, None, None], result)
+
+ def _validate_app_tenant(self, *, app_id: str, tenant_id: str) -> None:
+ with self._session_factory() as session:
+ app = session.get(App, app_id)
+ if app is None:
+ raise AgentLLMInnerServiceError(
+ "app_not_found",
+ "App not found.",
+ status_code=404,
+ )
+ if app.tenant_id != tenant_id:
+ raise AgentLLMInnerServiceError(
+ "app_tenant_mismatch",
+ "App does not belong to the caller tenant.",
+ status_code=403,
+ )
+
+
+__all__ = ["AgentLLMInnerService", "AgentLLMInnerServiceError", "PreparedAgentLLMInvocation"]
diff --git a/api/services/credit_pool_service.py b/api/services/credit_pool_service.py
index 2a18e52a34c..2e013a068ed 100644
--- a/api/services/credit_pool_service.py
+++ b/api/services/credit_pool_service.py
@@ -6,7 +6,7 @@ from piling up database transactions while preserving cross-tenant concurrency.
"""
import logging
-from collections.abc import Callable
+from collections.abc import Callable, Mapping
from dataclasses import dataclass, field
from enum import StrEnum, auto
from typing import Any
@@ -347,6 +347,8 @@ class CreditPoolService:
credits_required: int,
pool_type: str | ProviderQuotaType = "trial",
*,
+ request_id: str | None = None,
+ metadata: Mapping[str, str] | None = None,
session: Session | None = None,
) -> int:
"""Deduct exactly the requested credits or raise without mutating the pool."""
@@ -357,14 +359,15 @@ class CreditPoolService:
if cls._use_billing_quota():
from services.billing_service import BillingService
- request_id = str(uuid4())
+ resolved_request_id = request_id or str(uuid4())
+ billing_metadata = {"source": "credit_pool.check_and_deduct", **dict(metadata or {})}
result = BillingService.quota_reserve(
tenant_id=tenant_id,
feature_key=FEATURE_KEY_CREDIT_POOL,
bucket=normalized_pool_type,
- request_id=request_id,
+ request_id=resolved_request_id,
amount=credits_required,
- meta={"source": "credit_pool.check_and_deduct"},
+ meta=billing_metadata,
)
reservation_id = result.get("reservation_id", "")
if not reservation_id:
@@ -376,7 +379,7 @@ class CreditPoolService:
bucket=normalized_pool_type,
reservation_id=reservation_id,
actual_amount=credits_required,
- meta={"source": "credit_pool.check_and_deduct"},
+ meta=billing_metadata,
)
except Exception:
try:
diff --git a/api/services/entities/agent_llm_inner.py b/api/services/entities/agent_llm_inner.py
new file mode 100644
index 00000000000..55e6aba10df
--- /dev/null
+++ b/api/services/entities/agent_llm_inner.py
@@ -0,0 +1,66 @@
+"""DTOs for the API-owned Agent LLM gateway."""
+
+from __future__ import annotations
+
+from typing import ClassVar, Literal
+
+from pydantic import BaseModel, ConfigDict, Field, JsonValue
+
+from graphon.model_runtime.entities.message_entities import PromptMessageTool
+
+type AgentLLMMode = Literal["workflow_run", "single_step", "agent_app", "babysit", "fasten"]
+type AgentConfigVersionKind = Literal["snapshot", "draft", "build_draft"]
+
+
+class AgentLLMInvokeCaller(BaseModel):
+ invocation_id: str
+ agent_run_id: str
+ call_index: int = Field(ge=1)
+ tenant_id: str
+ user_id: str
+ user_from: Literal["account", "end-user"]
+ app_id: str
+ invoke_from: str
+ agent_mode: AgentLLMMode
+ conversation_id: str | None = None
+ workflow_id: str | None = None
+ workflow_run_id: str | None = None
+ node_id: str | None = None
+ node_execution_id: str | None = None
+ agent_id: str | None = None
+ agent_config_version_id: str | None = None
+ agent_config_version_kind: AgentConfigVersionKind | None = None
+ trace_id: str | None = None
+
+ model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid")
+
+
+class AgentLLMInvokeTarget(BaseModel):
+ provider: str
+ model: str
+ # The trusted Agent service already produces plugin-runtime prompt payloads.
+ # Keep them opaque here so this transport cannot discard role-specific or
+ # newly introduced fields before the authoritative runtime validates them.
+ prompt_messages: list[dict[str, JsonValue]]
+ model_parameters: dict[str, JsonValue] = Field(default_factory=dict)
+ tools: list[PromptMessageTool] | None = None
+ stop: list[str] | None = None
+ stream: bool = True
+
+ model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid", arbitrary_types_allowed=True)
+
+
+class AgentLLMInvokeRequest(BaseModel):
+ caller: AgentLLMInvokeCaller
+ target: AgentLLMInvokeTarget
+
+ model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid", arbitrary_types_allowed=True)
+
+
+__all__ = [
+ "AgentConfigVersionKind",
+ "AgentLLMInvokeCaller",
+ "AgentLLMInvokeRequest",
+ "AgentLLMInvokeTarget",
+ "AgentLLMMode",
+]
diff --git a/api/tests/unit_tests/clients/agent_backend/test_request_builder.py b/api/tests/unit_tests/clients/agent_backend/test_request_builder.py
index 6f2a10452c5..4f259ade3d6 100644
--- a/api/tests/unit_tests/clients/agent_backend/test_request_builder.py
+++ b/api/tests/unit_tests/clients/agent_backend/test_request_builder.py
@@ -53,7 +53,6 @@ def _run_input() -> AgentBackendWorkflowNodeRunInput:
plugin_id="langgenius/openai",
model_provider="openai",
model="gpt-test",
- credentials={"api_key": "secret-key"},
),
execution_context=DifyExecutionContextLayerConfig(
tenant_id="tenant-1",
@@ -126,7 +125,6 @@ def test_request_builder_forwards_plugin_specific_model_settings_via_extra_body(
plugin_id="langgenius/tongyi",
model_provider="tongyi",
model="qwen-plus-latest",
- credentials={"api_key": "secret-key"},
model_settings={"temperature": 0.7, "enable_thinking": True, "thinking_budget": 4096},
)
}
@@ -207,7 +205,9 @@ def test_request_builder_sets_model_and_output_layer_contract_ids():
assert execution_context_config.invoke_from == "debugger"
assert layers[DIFY_AGENT_HISTORY_LAYER_ID].type == PYDANTIC_AI_HISTORY_LAYER_TYPE_ID
assert layers[DIFY_AGENT_MODEL_LAYER_ID].type == DIFY_PLUGIN_LLM_LAYER_TYPE_ID
- assert cast(DifyPluginLLMLayerConfig, layers[DIFY_AGENT_MODEL_LAYER_ID].config).plugin_id == "langgenius/openai"
+ model_config = cast(DifyPluginLLMLayerConfig, layers[DIFY_AGENT_MODEL_LAYER_ID].config)
+ assert model_config.plugin_id == "langgenius/openai"
+ assert "credentials" not in model_config.model_dump(mode="json")
assert layers[DIFY_AGENT_MODEL_LAYER_ID].deps == {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID}
assert layers[DIFY_AGENT_OUTPUT_LAYER_ID].type == DIFY_OUTPUT_LAYER_TYPE_ID
@@ -309,12 +309,12 @@ def test_request_builder_rejects_blank_prompts():
)
-def test_redact_for_agent_backend_log_hides_credentials():
+def test_agent_backend_request_does_not_contain_llm_credentials():
request = AgentBackendRunRequestBuilder().build_for_workflow_node(_run_input())
redacted = cast(dict[str, Any], redact_for_agent_backend_log(request))
- assert redacted["composition"]["layers"][5]["config"]["credentials"] == "[REDACTED]"
+ assert "credentials" not in redacted["composition"]["layers"][5]["config"]
def _agent_app_input(*, include_shell: bool = False) -> AgentBackendAgentAppRunInput:
@@ -323,7 +323,6 @@ def _agent_app_input(*, include_shell: bool = False) -> AgentBackendAgentAppRunI
plugin_id="langgenius/openai",
model_provider="openai",
model="gpt-test",
- credentials={"api_key": "secret-key"},
),
execution_context=DifyExecutionContextLayerConfig(
tenant_id="tenant-1",
@@ -466,7 +465,6 @@ def test_agent_app_request_builder_adds_knowledge_layer_when_configured():
def test_ask_human_layer_injected_when_configured():
-
from dify_agent.layers.ask_human import DIFY_ASK_HUMAN_LAYER_TYPE_ID, DifyAskHumanLayerConfig
from clients.agent_backend.request_builder import DIFY_ASK_HUMAN_LAYER_ID
diff --git a/api/tests/unit_tests/controllers/inner_api/test_agent_llm.py b/api/tests/unit_tests/controllers/inner_api/test_agent_llm.py
new file mode 100644
index 00000000000..15160f8e749
--- /dev/null
+++ b/api/tests/unit_tests/controllers/inner_api/test_agent_llm.py
@@ -0,0 +1,227 @@
+"""Contract tests for the trusted Agent LLM streaming gateway."""
+
+import json
+from collections.abc import Generator
+from contextlib import contextmanager
+from unittest.mock import MagicMock, patch
+from uuid import uuid4
+
+from flask import Flask
+from pydantic import JsonValue
+
+from controllers.inner_api import bp as inner_api_bp
+from core.errors.error import QuotaExceededError
+from graphon.model_runtime.entities.llm_entities import LLMResultChunk, LLMResultChunkDelta
+from graphon.model_runtime.entities.message_entities import AssistantPromptMessage, UserPromptMessage
+from services.agent_llm_inner_service import AgentLLMInnerServiceError, PreparedAgentLLMInvocation
+from services.entities.agent_llm_inner import AgentLLMInvokeRequest
+
+
+def _payload() -> dict[str, object]:
+ return {
+ "caller": {
+ "invocation_id": str(uuid4()),
+ "agent_run_id": str(uuid4()),
+ "call_index": 1,
+ "tenant_id": str(uuid4()),
+ "user_id": str(uuid4()),
+ "user_from": "account",
+ "app_id": str(uuid4()),
+ "invoke_from": "debugger",
+ "agent_mode": "workflow_run",
+ "agent_config_version_kind": "draft",
+ },
+ "target": {
+ "provider": "openai",
+ "model": "gpt-test",
+ "prompt_messages": [UserPromptMessage(content="hello").model_dump(mode="json")],
+ "model_parameters": {"temperature": 0.2},
+ "stream": True,
+ },
+ }
+
+
+@contextmanager
+def _agent_inner_auth() -> Generator[None]:
+ with (
+ patch("configs.dify_config.PLUGIN_DAEMON_KEY", "plugin-daemon-key"),
+ patch("configs.dify_config.INNER_API_KEY_FOR_PLUGIN", "inner-key"),
+ ):
+ yield
+
+
+def _app() -> Flask:
+ app = Flask(__name__)
+ app.config["TESTING"] = True
+ app.register_blueprint(inner_api_bp)
+ return app
+
+
+def test_post_streams_plugin_compatible_envelope() -> None:
+ payload = _payload()
+ request = AgentLLMInvokeRequest.model_validate(payload)
+ prepared = PreparedAgentLLMInvocation(request=request, model_instance=MagicMock())
+ chunk = LLMResultChunk(
+ model="gpt-test",
+ delta=LLMResultChunkDelta(
+ index=0,
+ message=AssistantPromptMessage(content="done", tool_calls=[]),
+ ),
+ )
+
+ with (
+ _agent_inner_auth(),
+ patch("controllers.inner_api.agent.llm.AgentLLMInnerService.prepare", return_value=prepared) as prepare,
+ patch("controllers.inner_api.agent.llm.AgentLLMInnerService.invoke", return_value=iter([chunk])),
+ ):
+ response = (
+ _app()
+ .test_client()
+ .post(
+ "/inner/api/agent/llm/invoke",
+ json=payload,
+ headers={"X-Inner-Api-Key": "inner-key"},
+ )
+ )
+ assert response.status_code == 200
+ assert response.content_type == "text/event-stream"
+ data_line = response.get_data(as_text=True).strip().removeprefix("data: ")
+ envelope = json.loads(data_line)
+ assert envelope["code"] == 0
+ assert envelope["data"]["delta"]["message"]["content"] == "done"
+ prepare.assert_called_once()
+
+
+def test_post_preserves_prompt_messages_without_transport_validation() -> None:
+ payload = _payload()
+ prompt_messages: list[dict[str, JsonValue]] = [
+ {
+ "role": "assistant",
+ "content": "prior answer",
+ "tool_calls": [],
+ },
+ {
+ "role": "assistant",
+ "content": "",
+ "tool_calls": [
+ {
+ "id": "call-1",
+ "type": "function",
+ "function": {"name": "search", "arguments": '{"query":"Dify"}'},
+ }
+ ],
+ "provider_extension": {"thought_signature": "opaque"},
+ },
+ {
+ "role": "tool",
+ "content": "result",
+ "name": "search",
+ "tool_call_id": "call-1",
+ },
+ ]
+ target = payload["target"]
+ assert isinstance(target, dict)
+ target["prompt_messages"] = prompt_messages
+ captured: dict[str, AgentLLMInvokeRequest] = {}
+
+ def prepare(request: AgentLLMInvokeRequest) -> PreparedAgentLLMInvocation:
+ captured["request"] = request
+ return PreparedAgentLLMInvocation(request=request, model_instance=MagicMock())
+
+ with (
+ _agent_inner_auth(),
+ patch("controllers.inner_api.agent.llm.AgentLLMInnerService.prepare", side_effect=prepare),
+ patch("controllers.inner_api.agent.llm.AgentLLMInnerService.invoke", return_value=iter([])),
+ ):
+ response = (
+ _app()
+ .test_client()
+ .post(
+ "/inner/api/agent/llm/invoke",
+ json=payload,
+ headers={"X-Inner-Api-Key": "inner-key"},
+ )
+ )
+
+ assert response.status_code == 200
+ assert captured["request"].target.prompt_messages == prompt_messages
+
+
+def test_post_rejects_invalid_body_before_model_resolution() -> None:
+ with _agent_inner_auth():
+ response = (
+ _app()
+ .test_client()
+ .post(
+ "/inner/api/agent/llm/invoke",
+ json={"caller": {}},
+ headers={"X-Inner-Api-Key": "inner-key"},
+ )
+ )
+
+ assert response.status_code == 400
+ assert response.get_json()["code"] == "invalid_request"
+
+
+def test_post_preserves_preflight_quota_failure() -> None:
+ with (
+ _agent_inner_auth(),
+ patch(
+ "controllers.inner_api.agent.llm.AgentLLMInnerService.prepare",
+ side_effect=AgentLLMInnerServiceError(
+ "agent_llm_quota_exceeded",
+ "Insufficient Message Credits.",
+ status_code=429,
+ ),
+ ),
+ ):
+ response = (
+ _app()
+ .test_client()
+ .post(
+ "/inner/api/agent/llm/invoke",
+ json=_payload(),
+ headers={"X-Inner-Api-Key": "inner-key"},
+ )
+ )
+
+ assert response.status_code == 429
+ assert response.get_json() == {
+ "code": "agent_llm_quota_exceeded",
+ "message": "Insufficient Message Credits.",
+ "status": 429,
+ }
+
+
+def test_post_preserves_stream_quota_failure() -> None:
+ payload = _payload()
+ prepared = PreparedAgentLLMInvocation(
+ request=AgentLLMInvokeRequest.model_validate(payload),
+ model_instance=MagicMock(),
+ )
+ with (
+ _agent_inner_auth(),
+ patch("controllers.inner_api.agent.llm.AgentLLMInnerService.prepare", return_value=prepared),
+ patch(
+ "controllers.inner_api.agent.llm.AgentLLMInnerService.invoke",
+ side_effect=QuotaExceededError("Insufficient hosted model quota remaining."),
+ ),
+ ):
+ response = (
+ _app()
+ .test_client()
+ .post(
+ "/inner/api/agent/llm/invoke",
+ json=payload,
+ headers={"X-Inner-Api-Key": "inner-key"},
+ )
+ )
+
+ envelope = json.loads(response.get_data(as_text=True).strip().removeprefix("data: "))
+ error = json.loads(envelope["message"])
+ assert response.status_code == 200
+ assert envelope["code"] == -429
+ assert error == {
+ "error_type": "AgentLLMQuotaExceededError",
+ "message": "Insufficient hosted model quota remaining.",
+ }
diff --git a/api/tests/unit_tests/core/app/apps/agent_app/test_app_generator.py b/api/tests/unit_tests/core/app/apps/agent_app/test_app_generator.py
index d346dfe9601..ca0e989d9d1 100644
--- a/api/tests/unit_tests/core/app/apps/agent_app/test_app_generator.py
+++ b/api/tests/unit_tests/core/app/apps/agent_app/test_app_generator.py
@@ -274,7 +274,6 @@ class TestGenerateWorker:
mocker.patch(f"{MODULE}.session_factory.create_session", return_value=resolver_context)
mocker.patch(f"{MODULE}.db.session.close")
mocker.patch(f"{MODULE}.DifyRunContext", return_value=mocker.MagicMock())
- mocker.patch(f"{MODULE}.build_dify_model_access", return_value=(mocker.MagicMock(), None))
mocker.patch(f"{MODULE}.AgentAppRuntimeRequestBuilder", return_value=mocker.MagicMock())
mocker.patch(f"{MODULE}.create_agent_backend_run_client", return_value=mocker.MagicMock())
mocker.patch(f"{MODULE}.AgentBackendRunEventAdapter", return_value=mocker.MagicMock())
diff --git a/api/tests/unit_tests/core/app/apps/agent_app/test_app_runner.py b/api/tests/unit_tests/core/app/apps/agent_app/test_app_runner.py
index 26e44377e3a..b351f49c37f 100644
--- a/api/tests/unit_tests/core/app/apps/agent_app/test_app_runner.py
+++ b/api/tests/unit_tests/core/app/apps/agent_app/test_app_runner.py
@@ -77,11 +77,6 @@ def _thought_rows(session: Session) -> list[MessageAgentThought]:
return list(session.scalars(select(MessageAgentThought).order_by(MessageAgentThought.position)).all())
-class _FakeCredentialsProvider:
- def fetch(self, provider_name: str, model_name: str) -> dict[str, Any]:
- return {"openai_api_key": "sk-test"}
-
-
class _NoToolsBuilder:
def build_layers(self, **kwargs: Any) -> WorkflowAgentToolLayers:
del kwargs
@@ -488,7 +483,6 @@ def _runner(
) -> AgentAppRunner:
return AgentAppRunner(
request_builder=AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
),
agent_backend_client=client,
diff --git a/api/tests/unit_tests/core/app/apps/agent_app/test_runtime_request_builder.py b/api/tests/unit_tests/core/app/apps/agent_app/test_runtime_request_builder.py
index 1dc9d6933a8..7c06cc400d5 100644
--- a/api/tests/unit_tests/core/app/apps/agent_app/test_runtime_request_builder.py
+++ b/api/tests/unit_tests/core/app/apps/agent_app/test_runtime_request_builder.py
@@ -4,7 +4,6 @@
from __future__ import annotations
from types import SimpleNamespace
-from typing import Any
import pytest
from dify_agent.layers.dify_core_tools import DifyCoreToolConfig, DifyCoreToolsLayerConfig
@@ -82,11 +81,6 @@ class TestBuildForAgentApp:
assert [layer.name for layer in request.composition.layers][0] == "agent_app_user_prompt"
-class _FakeCredentialsProvider:
- def fetch(self, provider_name: str, model_name: str) -> dict[str, Any]:
- return {"openai_api_key": "sk-test", "max": 5}
-
-
class _NoToolsBuilder:
def build_layers(self, **kwargs):
del kwargs
@@ -183,7 +177,6 @@ def _soul_with_model() -> AgentSoulConfig:
class TestAgentAppRuntimeRequestBuilder:
def test_build_maps_soul_to_run_request(self):
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
result = builder.build(_ctx(_soul_with_model()))
@@ -212,13 +205,12 @@ class TestAgentAppRuntimeRequestBuilder:
assert exec_ctx.config.invoke_from == "web-app"
assert exec_ctx.config.agent_mode == "agent_app"
assert req.on_exit.default.value == "suspend"
- # credentials are redacted in the log-safe view.
- assert result.redacted_request["composition"]["layers"][-1]["config"]["credentials"] == "[REDACTED]"
+ # LLM credentials are resolved by API and never enter the Agent request.
+ assert "credentials" not in result.redacted_request["composition"]["layers"][-1]["config"]
assert result.metadata["conversation_id"] == "conv-1"
def test_build_wraps_agent_soul_prompt_for_build_draft(self):
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -231,7 +223,6 @@ class TestAgentAppRuntimeRequestBuilder:
def test_build_propagates_draft_version_kind_without_wrapping_prompt(self):
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -258,7 +249,6 @@ class TestAgentAppRuntimeRequestBuilder:
]
tools_builder = _PluginLayerBuilder()
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=tools_builder, # type: ignore[arg-type]
)
@@ -279,7 +269,6 @@ class TestAgentAppRuntimeRequestBuilder:
]
tools_builder = _PluginLayerBuilder()
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=tools_builder, # type: ignore[arg-type]
)
@@ -299,7 +288,6 @@ class TestAgentAppRuntimeRequestBuilder:
}
]
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_CoreLayerBuilder(), # type: ignore[arg-type]
)
@@ -315,7 +303,6 @@ class TestAgentAppRuntimeRequestBuilder:
"langgenius/openai:0.4.2@21195ee1321849e0a7d4b3f6b2fd8c2be23ea6c7182e1b444ecc4c1711b52468"
)
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -329,7 +316,6 @@ class TestAgentAppRuntimeRequestBuilder:
soul = _soul_with_model()
soul.model.plugin_id = "langgenius/openai/openai"
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -365,7 +351,6 @@ class TestAgentAppRuntimeRequestBuilder:
}
)
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -384,7 +369,6 @@ class TestAgentAppRuntimeRequestBuilder:
def test_build_raises_when_model_missing(self):
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
with pytest.raises(AgentAppRuntimeRequestBuildError) as exc:
@@ -406,7 +390,6 @@ class TestAgentAppRuntimeRequestBuilder:
}
)
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -448,7 +431,6 @@ def _soul_with_model_and_skill() -> AgentSoulConfig:
class TestAgentAppConfigLayer:
def test_config_layer_injected(self):
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -475,7 +457,6 @@ class TestAgentAppConfigLayer:
def test_config_layer_present_when_agent_soul_has_no_config_assets(self, monkeypatch: pytest.MonkeyPatch):
monkeypatch.setattr("core.app.apps.agent_app.runtime_request_builder.dify_config.AGENT_SHELL_ENABLED", True)
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -500,7 +481,6 @@ class TestAgentAppConfigLayer:
def test_config_layer_for_build_draft_marks_config_writable(self):
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -546,7 +526,6 @@ class TestAgentAppConfigLayer:
soul = _soul_with_model_and_skill()
soul.prompt.system_prompt = system_prompt
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
@@ -564,7 +543,6 @@ class TestAgentAppConfigLayer:
"Use [§skill:ghost-skill:Ghost Skill§], [§file:ghost.txt:Ghost File§], and [§file:no-label.txt§]."
)
builder = AgentAppRuntimeRequestBuilder(
- credentials_provider=_FakeCredentialsProvider(),
dify_tools_builder=_NoToolsBuilder(), # type: ignore[arg-type]
)
diff --git a/api/tests/unit_tests/core/app/test_llm_quota.py b/api/tests/unit_tests/core/app/test_llm_quota.py
index 64372ae91d2..55e4293dae9 100644
--- a/api/tests/unit_tests/core/app/test_llm_quota.py
+++ b/api/tests/unit_tests/core/app/test_llm_quota.py
@@ -2,6 +2,7 @@ from collections.abc import Generator
from contextlib import contextmanager
from types import SimpleNamespace
from unittest.mock import ANY, MagicMock, patch
+from uuid import UUID
import pytest
from sqlalchemy import create_engine, select
@@ -131,6 +132,7 @@ def test_reserve_llm_quota_uses_exact_credit_pool_reservation() -> None:
tenant_id="tenant-id",
provider="openai",
model="gpt-4o",
+ request_id="11111111-1111-5111-8111-111111111111",
)
reservation.commit(LLMUsage.empty_usage())
reservation.release()
@@ -141,7 +143,7 @@ def test_reserve_llm_quota_uses_exact_credit_pool_reservation() -> None:
tenant_id="tenant-id",
credits_required=9,
pool_type="trial",
- request_id=ANY,
+ request_id="11111111-1111-5111-8111-111111111111",
session_factory=ANY,
meta={"source": "llm.invoke", "provider": "openai", "model": "gpt-4o"},
)
@@ -149,6 +151,35 @@ def test_reserve_llm_quota_uses_exact_credit_pool_reservation() -> None:
credit_reservation.release.assert_not_called()
+def test_reserve_llm_quota_generates_request_id_when_not_supplied() -> None:
+ credit_reservation = MagicMock()
+ provider_configuration = SimpleNamespace(
+ using_provider_type=ProviderType.SYSTEM,
+ get_provider_model=MagicMock(return_value=SimpleNamespace(status=ModelStatus.ACTIVE)),
+ system_configuration=SimpleNamespace(
+ current_quota_type=ProviderQuotaType.TRIAL,
+ quota_configurations=[
+ SimpleNamespace(
+ quota_type=ProviderQuotaType.TRIAL,
+ quota_unit=QuotaUnit.TIMES,
+ quota_limit=100,
+ )
+ ],
+ ),
+ )
+ provider_manager = MagicMock()
+ provider_manager.get_configurations.return_value.get.return_value = provider_configuration
+
+ with (
+ patch("core.app.llm.quota.create_plugin_provider_manager", return_value=provider_manager),
+ patch("core.app.llm.quota.CreditPoolService.reserve_credits", return_value=credit_reservation) as reserve,
+ ):
+ reserve_llm_quota_for_model(tenant_id="tenant-id", provider="openai", model="gpt-4o")
+
+ generated_request_id = reserve.call_args.kwargs["request_id"]
+ assert str(UUID(generated_request_id)) == generated_request_id
+
+
def test_reserve_non_llm_quota_uses_model_type_and_credit_pool_reservation() -> None:
credit_reservation = MagicMock()
provider_configuration = SimpleNamespace(
diff --git a/api/tests/unit_tests/core/test_model_manager.py b/api/tests/unit_tests/core/test_model_manager.py
index 398d6c4f158..18c5245e851 100644
--- a/api/tests/unit_tests/core/test_model_manager.py
+++ b/api/tests/unit_tests/core/test_model_manager.py
@@ -1,6 +1,7 @@
from collections.abc import Callable
from io import BytesIO
from unittest.mock import MagicMock, patch
+from uuid import uuid4
import pytest
import redis
@@ -175,13 +176,19 @@ def test_quota_managed_non_streaming_invocation_finalizes_reservation() -> None:
result = MagicMock(spec=LLMResult, usage=usage)
reservation = MagicMock(commit_before_delivery=True)
+ invocation_id = str(uuid4())
with (
- patch.object(model_instance, "reserve_quota", return_value=reservation),
+ patch.object(model_instance, "reserve_quota", return_value=reservation) as reserve_quota,
patch.object(ModelInstance, "invoke_llm", return_value=result) as invoke,
):
- response = model_instance.invoke_llm(prompt_messages=[], stream=False)
+ response = model_instance.invoke_llm(
+ prompt_messages=[],
+ stream=False,
+ request_metadata={"invocation_id": invocation_id},
+ )
assert response is result
+ reserve_quota.assert_called_once_with(request_id=invocation_id)
invoke.assert_called_once()
reservation.commit.assert_called_once_with(usage)
reservation.release.assert_called_once_with()
@@ -202,17 +209,23 @@ def test_quota_managed_stream_commits_before_first_chunk() -> None:
events: list[str] = []
reservation.commit.side_effect = lambda _usage: events.append("commit")
+ invocation_id = str(uuid4())
with (
- patch.object(model_instance, "reserve_quota", return_value=reservation),
+ patch.object(model_instance, "reserve_quota", return_value=reservation) as reserve_quota,
patch.object(ModelInstance, "invoke_llm", return_value=(item for item in [chunk])),
):
- response = model_instance.invoke_llm(prompt_messages=[], stream=True)
+ response = model_instance.invoke_llm(
+ prompt_messages=[],
+ stream=True,
+ request_metadata={"invocation_id": invocation_id},
+ )
assert next(response) is chunk
events.append("delivered")
with pytest.raises(StopIteration):
next(response)
assert events == ["commit", "delivered"]
+ reserve_quota.assert_called_once_with(request_id=invocation_id)
reservation.release.assert_called_once_with()
diff --git a/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_agent_node.py b/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_agent_node.py
index 1ee031d9462..915cdaa8ec1 100644
--- a/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_agent_node.py
+++ b/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_agent_node.py
@@ -62,13 +62,6 @@ from models.agent_config_entities import (
from services.agent.workspace_service import AgentWorkspaceNotFoundError
-class FakeCredentialsProvider:
- def fetch(self, provider_name: str, model_name: str) -> dict[str, object]:
- assert provider_name == "openai"
- assert model_name == "gpt-test"
- return {"api_key": "secret-key"}
-
-
def _restored_file(*, transfer_method: FileTransferMethod, reference: str) -> File:
return File(
type=FileType.DOCUMENT,
@@ -360,8 +353,7 @@ def _node(
),
),
binding_resolver=binding_resolver,
- runtime_request_builder=runtime_request_builder
- or WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()),
+ runtime_request_builder=runtime_request_builder or WorkflowAgentRuntimeRequestBuilder(),
agent_backend_client=client,
event_adapter=AgentBackendRunEventAdapter(),
output_adapter=WorkflowAgentOutputAdapter(),
@@ -402,7 +394,7 @@ def test_agent_node_run_maps_successful_agent_backend_run_to_node_result():
assert agent_log["agent_backend"]["status"] == "succeeded"
assert result.process_data["agent_id"] == "agent-1"
layers = {layer["name"]: layer for layer in result.inputs["agent_backend_request"]["composition"]["layers"]}
- assert layers["llm"]["config"]["credentials"] == "[REDACTED]"
+ assert "credentials" not in layers["llm"]["config"]
def test_agent_node_uses_resolved_backend_binding_before_backend_invocation() -> None:
@@ -466,7 +458,7 @@ def test_agent_node_maps_persisted_participant_lookup_error_to_node_failure() ->
def test_agent_node_passes_execution_id_to_session_store_and_runtime_request_builder() -> None:
store = FakeSessionStore()
- request_builder = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider())
+ request_builder = WorkflowAgentRuntimeRequestBuilder()
node = _node(session_store=store, runtime_request_builder=request_builder)
execution_id = node.execution_id
diff --git a/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_runtime_request_builder.py b/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_runtime_request_builder.py
index bd2302e6a6a..2d2e25ca919 100644
--- a/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_runtime_request_builder.py
+++ b/api/tests/unit_tests/core/workflow/nodes/agent_v2/test_runtime_request_builder.py
@@ -40,24 +40,6 @@ from models.agent_config_entities import (
)
-class FakeCredentialsProvider:
- def fetch(self, provider_name: str, model_name: str) -> dict[str, object]:
- assert provider_name == "openai"
- assert model_name == "gpt-test"
- return {"api_key": "secret-key"}
-
-
-class CapturingCredentialsProvider:
- def __init__(self) -> None:
- self.provider_name: str | None = None
- self.model_name: str | None = None
-
- def fetch(self, provider_name: str, model_name: str) -> dict[str, object]:
- self.provider_name = provider_name
- self.model_name = model_name
- return {"api_key": "secret-key"}
-
-
def test_agent_soul_round_trip_preserves_existing_app_feature_fields():
config = AgentSoulConfig.model_validate(
{
@@ -239,7 +221,7 @@ def _uploaded_workflow_files_prompt_payload(result) -> object:
def test_builds_create_run_request_from_agent_soul_and_node_job():
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(_context())
+ result = WorkflowAgentRuntimeRequestBuilder().build(_context())
dumped = result.request.model_dump(mode="json")
layers = {layer["name"]: layer for layer in dumped["composition"]["layers"]}
@@ -257,7 +239,7 @@ def test_builds_create_run_request_from_agent_soul_and_node_job():
assert dumped["composition"]["layers"][-1]["config"]["json_schema"]["properties"]["summary"]["type"] == "string"
assert DIFY_AGENT_HISTORY_LAYER_ID in layers
redacted_layers = {layer["name"]: layer for layer in result.redacted_request["composition"]["layers"]}
- assert redacted_layers[DIFY_AGENT_MODEL_LAYER_ID]["config"]["credentials"] == "[REDACTED]"
+ assert "credentials" not in redacted_layers[DIFY_AGENT_MODEL_LAYER_ID]["config"]
def test_build_includes_plugin_tools_layer_returned_by_injected_builder_for_debugger():
@@ -277,7 +259,6 @@ def test_build_includes_plugin_tools_layer_returned_by_injected_builder_for_debu
)
result = WorkflowAgentRuntimeRequestBuilder(
- credentials_provider=FakeCredentialsProvider(),
dify_tools_builder=tools_builder, # type: ignore[arg-type]
).build(context)
@@ -308,7 +289,6 @@ def test_build_forwards_service_api_invoke_from_to_injected_plugin_layer_builder
)
result = WorkflowAgentRuntimeRequestBuilder(
- credentials_provider=FakeCredentialsProvider(),
dify_tools_builder=tools_builder, # type: ignore[arg-type]
).build(context)
@@ -334,7 +314,6 @@ def test_build_includes_core_tools_layer_returned_by_injected_builder():
)
result = WorkflowAgentRuntimeRequestBuilder(
- credentials_provider=FakeCredentialsProvider(),
dify_tools_builder=FakeCoreLayerBuilder(), # type: ignore[arg-type]
).build(context)
@@ -363,15 +342,11 @@ def test_normalizes_langgenius_model_provider_for_agent_backend_transport(plugin
model="gpt-test",
),
)
- credentials_provider = CapturingCredentialsProvider()
-
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=credentials_provider).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
layers = {layer["name"]: layer for layer in dumped["composition"]["layers"]}
model_config = layers[DIFY_AGENT_MODEL_LAYER_ID]["config"]
- assert credentials_provider.provider_name == "langgenius/openai/openai"
- assert credentials_provider.model_name == "gpt-test"
assert model_config["plugin_id"] == "langgenius/openai"
assert model_config["model_provider"] == "openai"
@@ -414,7 +389,7 @@ def test_builds_workflow_run_request_with_file_output_schema_and_reserved_metada
dify_context = context.dify_context.model_copy(update={"invoke_from": InvokeFrom.SERVICE_API})
context = replace(context, dify_context=dify_context, workflow_run_id=None, snapshot=snapshot, binding=binding)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
layers = {layer["name"]: layer for layer in dumped["composition"]["layers"]}
@@ -473,7 +448,7 @@ def test_build_maps_agent_soul_shell_settings_to_shell_layer(monkeypatch: pytest
)
context = replace(context, snapshot=snapshot)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
shell_config = {layer["name"]: layer for layer in dumped["composition"]["layers"]}[DIFY_SHELL_LAYER_ID]["config"]
@@ -691,7 +666,6 @@ def test_builds_workflow_run_request_with_dify_plugin_tools_layer(monkeypatch: p
dify_tools_builder = CapturingPluginLayerBuilder()
result = WorkflowAgentRuntimeRequestBuilder(
- credentials_provider=FakeCredentialsProvider(),
dify_tools_builder=cast(WorkflowAgentDifyToolsBuilder, dify_tools_builder),
).build(context)
@@ -796,7 +770,7 @@ def test_build_maps_agent_soul_knowledge_to_knowledge_layer_config():
)
context = replace(context, snapshot=snapshot)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
layers = {layer["name"]: layer for layer in dumped["composition"]["layers"]}
@@ -907,7 +881,7 @@ def test_build_knowledge_layer_maps_disabled_score_threshold_to_zero():
)
context = replace(context, snapshot=snapshot)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
knowledge_layer = next(layer for layer in dumped["composition"]["layers"] if layer["name"] == "knowledge")
@@ -935,7 +909,7 @@ def test_build_skips_knowledge_layer_when_agent_soul_has_no_sets():
)
context = replace(context, snapshot=snapshot)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
assert all(layer["name"] != "knowledge" for layer in dumped["composition"]["layers"])
@@ -945,7 +919,7 @@ def test_build_passes_saved_session_snapshot_to_agent_backend_request():
session_snapshot = CompositorSessionSnapshot(layers=[])
context = replace(_context(), session_snapshot=session_snapshot)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
assert result.request.session_snapshot is session_snapshot
@@ -962,7 +936,7 @@ def test_requires_agent_soul_model_config():
context = replace(context, snapshot=snapshot)
with pytest.raises(WorkflowAgentRuntimeRequestBuildError, match="Agent Soul model"):
- WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ WorkflowAgentRuntimeRequestBuilder().build(context)
def test_missing_previous_node_output_fails_request_build():
@@ -984,7 +958,7 @@ def test_missing_previous_node_output_fails_request_build():
context = replace(context, binding=binding)
with pytest.raises(WorkflowAgentRuntimeRequestBuildError) as exc_info:
- WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ WorkflowAgentRuntimeRequestBuilder().build(context)
assert exc_info.value.error_code == "missing_previous_node_output"
@@ -1008,7 +982,7 @@ def test_invalid_previous_node_output_ref_fails_request_build():
context = replace(context, binding=binding)
with pytest.raises(WorkflowAgentRuntimeRequestBuildError) as exc_info:
- WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ WorkflowAgentRuntimeRequestBuilder().build(context)
assert exc_info.value.error_code == "invalid_previous_node_output_ref"
@@ -1029,7 +1003,7 @@ def test_empty_declared_outputs_injects_prd_defaults_text_files_json():
)
context = replace(context, binding=binding)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
output_layer = dumped["composition"]["layers"][-1]["config"]
@@ -1074,7 +1048,7 @@ def test_array_output_emits_typed_items_per_array_item():
)
context = replace(context, binding=binding)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
output_schema = result.request.model_dump(mode="json")["composition"]["layers"][-1]["config"]["json_schema"]
tags_schema = output_schema["properties"]["tags"]
@@ -1147,7 +1121,7 @@ def test_mentions_expand_in_soul_and_job_prompts_without_token_leak():
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
layers = _request_layers(result)
agent_soul_prompt = layers["agent_soul_prompt"]["config"]["prefix"]
@@ -1191,7 +1165,7 @@ def test_previous_node_file_output_uses_agent_stub_download_mapping_in_workflow_
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
assert _request_layers(result)["workflow_node_job_prompt"]["config"]["user"] == (
"Review {{#previous-node.report#}} before responding."
@@ -1225,7 +1199,7 @@ def test_previous_node_file_mapping_strips_extra_fields_in_workflow_context():
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
assert _previous_node_prompt_payload(result, "previous-node.report") == {
"transfer_method": "tool_file",
@@ -1241,7 +1215,7 @@ def test_scalar_previous_node_output_appears_in_workflow_context_section():
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
user_prompt = _workflow_user_prompt(result)
@@ -1258,7 +1232,7 @@ def test_stale_previous_node_refs_are_ignored_when_workflow_prompt_has_no_fronte
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
assert _request_layers(result)["workflow_node_job_prompt"]["config"]["user"] == (
"Review the current request without upstream context."
@@ -1306,7 +1280,7 @@ def test_previous_node_file_array_uses_agent_stub_download_mappings_in_workflow_
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
assert _previous_node_prompt_payload(result, "previous-node.attachments") == [
{
@@ -1349,7 +1323,7 @@ def test_uploaded_workflow_files_are_included_without_prompt_marker():
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
user_prompt = _workflow_user_prompt(result)
assert "- Uploaded workflow files:" in user_prompt
@@ -1389,7 +1363,7 @@ def test_previous_node_remote_url_file_mapping_is_not_truncated_in_workflow_cont
}
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
assert _previous_node_prompt_payload(result, "previous-node.report") == {
"transfer_method": "remote_url",
@@ -1463,7 +1437,7 @@ def test_build_config_layer_config_returns_empty_config_for_empty_agent_soul():
def test_workflow_run_request_has_config_layer_with_empty_agent_soul(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setattr("core.workflow.nodes.agent_v2.runtime_request_builder.dify_config.AGENT_SHELL_ENABLED", True)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(_context())
+ result = WorkflowAgentRuntimeRequestBuilder().build(_context())
dumped = result.request.model_dump(mode="json")
layers = {layer["name"]: layer for layer in dumped["composition"]["layers"]}
@@ -1489,7 +1463,7 @@ def test_workflow_run_request_contains_config_layer():
context = _context()
context.snapshot.config_snapshot = _soul_with_config_assets()
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
dumped = result.request.model_dump(mode="json")
layer_names = [layer["name"] for layer in dumped["composition"]["layers"]]
@@ -1525,7 +1499,7 @@ def test_workflow_runtime_expands_config_mentions_in_agent_soul_prompt():
context = _context()
context.snapshot.config_snapshot = _soul_with_config_assets()
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
soul_prompt = next(layer for layer in result.request.composition.layers if layer.name == "agent_soul_prompt")
assert soul_prompt.config.prefix == "You are careful. Use tender-analyzer and sample.pdf."
@@ -1543,7 +1517,7 @@ def test_workflow_runtime_missing_config_mentions_fall_back_to_label_then_name()
model=AgentSoulModelConfig(plugin_id="langgenius/openai", model_provider="openai", model="gpt-test"),
)
- result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+ result = WorkflowAgentRuntimeRequestBuilder().build(context)
soul_prompt = next(layer for layer in result.request.composition.layers if layer.name == "agent_soul_prompt")
assert soul_prompt.config.prefix == "Use Ghost Skill, Ghost File, and no-label.txt."
diff --git a/api/tests/unit_tests/events/test_update_provider_when_message_created.py b/api/tests/unit_tests/events/test_update_provider_when_message_created.py
index a700783b31c..7c55c949fab 100644
--- a/api/tests/unit_tests/events/test_update_provider_when_message_created.py
+++ b/api/tests/unit_tests/events/test_update_provider_when_message_created.py
@@ -8,7 +8,7 @@ from sqlalchemy import select
from sqlalchemy.engine import Engine
from sqlalchemy.orm import Session, sessionmaker
-from core.app.entities.app_invoke_entities import ChatAppGenerateEntity
+from core.app.entities.app_invoke_entities import AgentAppGenerateEntity, ChatAppGenerateEntity
from core.entities.provider_entities import ProviderQuotaType, QuotaUnit
from events.event_handlers import update_provider_when_message_created
from models import Message, TenantCreditPool
@@ -121,6 +121,48 @@ def test_message_created_paid_credit_accounting_uses_paid_pool() -> None:
)
+def test_agent_app_gateway_accounting_skips_legacy_message_charge() -> None:
+ tenant_id = str(uuid4())
+ system_configuration = SimpleNamespace(
+ current_quota_type=ProviderQuotaType.TRIAL,
+ quota_configurations=[
+ SimpleNamespace(
+ quota_type=ProviderQuotaType.TRIAL,
+ quota_unit=QuotaUnit.CREDITS,
+ quota_limit=10,
+ )
+ ],
+ )
+ application_generate_entity = AgentAppGenerateEntity.model_construct(
+ app_config=SimpleNamespace(tenant_id=tenant_id),
+ model_conf=SimpleNamespace(
+ provider="openai",
+ model="gpt-4o",
+ provider_model_bundle=SimpleNamespace(
+ configuration=SimpleNamespace(
+ using_provider_type=ProviderType.SYSTEM,
+ system_configuration=system_configuration,
+ )
+ ),
+ ),
+ agent_llm_gateway_enabled=True,
+ )
+
+ with (
+ patch.object(update_provider_when_message_created, "_deduct_credit_pool_quota_capped") as mock_deduct,
+ patch.object(update_provider_when_message_created, "_execute_provider_updates") as mock_updates,
+ ):
+ update_provider_when_message_created.handle(
+ sender=Message(message_tokens=2, answer_tokens=1),
+ application_generate_entity=application_generate_entity,
+ )
+
+ mock_deduct.assert_not_called()
+ mock_updates.assert_called_once()
+ assert len(mock_updates.call_args.args[0]) == 1
+ assert mock_updates.call_args.args[0][0].description == "basic_last_used_update"
+
+
def test_capped_credit_pool_accounting_skips_exhaustion_warning_when_full_amount_is_deducted(
caplog: pytest.LogCaptureFixture,
) -> None:
diff --git a/api/tests/unit_tests/services/test_agent_llm_inner_service.py b/api/tests/unit_tests/services/test_agent_llm_inner_service.py
new file mode 100644
index 00000000000..de8fa2fea4f
--- /dev/null
+++ b/api/tests/unit_tests/services/test_agent_llm_inner_service.py
@@ -0,0 +1,348 @@
+"""Composition tests for the API-owned Agent LLM gateway."""
+
+from collections.abc import Generator
+from decimal import Decimal
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+from uuid import uuid4
+
+import pytest
+from pydantic import JsonValue
+from sqlalchemy.orm import Session, sessionmaker
+
+from configs import dify_config
+from core.entities.model_entities import ModelStatus
+from core.entities.provider_entities import ProviderQuotaType, QuotaUnit
+from core.model_manager import ModelInstance, QuotaManagedModelInstance
+from graphon.model_runtime.entities.llm_entities import LLMResultChunk, LLMResultChunkDelta, LLMUsage
+from graphon.model_runtime.entities.message_entities import AssistantPromptMessage, UserPromptMessage
+from models.model import App, AppMode
+from models.provider import ProviderType
+from services.agent_llm_inner_service import AgentLLMInnerService, AgentLLMInnerServiceError, PreparedAgentLLMInvocation
+from services.entities.agent_llm_inner import AgentLLMInvokeCaller, AgentLLMInvokeRequest, AgentLLMInvokeTarget
+
+
+def _request() -> AgentLLMInvokeRequest:
+ return AgentLLMInvokeRequest(
+ caller=AgentLLMInvokeCaller(
+ invocation_id=str(uuid4()),
+ agent_run_id=str(uuid4()),
+ call_index=1,
+ tenant_id=str(uuid4()),
+ user_id=str(uuid4()),
+ user_from="account",
+ app_id=str(uuid4()),
+ invoke_from="debugger",
+ agent_mode="workflow_run",
+ agent_config_version_kind="draft",
+ trace_id="trace-1",
+ ),
+ target=AgentLLMInvokeTarget(
+ provider="openai",
+ model="gpt-test",
+ prompt_messages=[UserPromptMessage(content="hello").model_dump(mode="json")],
+ ),
+ )
+
+
+def _model_instance(
+ *,
+ status: ModelStatus = ModelStatus.ACTIVE,
+) -> tuple[QuotaManagedModelInstance, MagicMock]:
+ provider_model = MagicMock()
+ provider_model.status = status
+ configuration = MagicMock()
+ configuration.get_provider_model.return_value = provider_model
+ bundle = MagicMock()
+ bundle.configuration = configuration
+
+ instance = object.__new__(QuotaManagedModelInstance)
+ instance.provider_model_bundle = bundle
+ instance.model_name = "gpt-test"
+ instance.provider = "openai"
+ instance.credentials = {"api_key": "hosted"}
+ instance.model_type_instance = MagicMock()
+ instance.load_balancing_manager = None
+ return instance, provider_model
+
+
+def _persist_app(
+ session: Session,
+ *,
+ request: AgentLLMInvokeRequest,
+ tenant_id: str | None = None,
+) -> App:
+ app = App(
+ id=request.caller.app_id,
+ tenant_id=tenant_id or request.caller.tenant_id,
+ name="Agent LLM gateway test app",
+ description="",
+ mode=AppMode.CHAT,
+ enable_site=False,
+ enable_api=False,
+ max_active_requests=None,
+ )
+ session.add(app)
+ session.commit()
+ return app
+
+
+def _prepare(
+ service: AgentLLMInnerService,
+ request: AgentLLMInvokeRequest,
+ model_instance: QuotaManagedModelInstance,
+) -> PreparedAgentLLMInvocation:
+ manager = MagicMock()
+ manager.get_model_instance.return_value = model_instance
+ with (
+ patch("services.agent_llm_inner_service.create_plugin_provider_manager"),
+ patch("services.agent_llm_inner_service.ModelManager", return_value=manager),
+ ):
+ return service.prepare(request)
+
+
+def _usage(*, total_tokens: int = 8) -> LLMUsage:
+ return LLMUsage(
+ prompt_tokens=3,
+ prompt_unit_price=Decimal(0),
+ prompt_price_unit=Decimal(0),
+ prompt_price=Decimal(0),
+ completion_tokens=total_tokens - 3,
+ completion_unit_price=Decimal(0),
+ completion_price_unit=Decimal(0),
+ completion_price=Decimal(0),
+ total_tokens=total_tokens,
+ total_price=Decimal(0),
+ currency="USD",
+ latency=0.1,
+ )
+
+
+def _chunk(text: str, *, usage: LLMUsage | None = None) -> LLMResultChunk:
+ return LLMResultChunk(
+ model="gpt-test",
+ delta=LLMResultChunkDelta(
+ index=0,
+ message=AssistantPromptMessage(content=text, tool_calls=[]),
+ usage=usage,
+ ),
+ )
+
+
+def test_prepare_rejects_missing_app(sqlite_session_factory: sessionmaker[Session]) -> None:
+ request = _request()
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+
+ with pytest.raises(AgentLLMInnerServiceError) as exc_info:
+ service.prepare(request)
+
+ assert exc_info.value.error_code == "app_not_found"
+ assert exc_info.value.status_code == 404
+
+
+def test_prepare_rejects_cross_tenant_app(
+ sqlite_session_factory: sessionmaker[Session],
+ sqlite_session: Session,
+) -> None:
+ request = _request()
+ _persist_app(sqlite_session, request=request, tenant_id=str(uuid4()))
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+
+ with pytest.raises(AgentLLMInnerServiceError) as exc_info:
+ service.prepare(request)
+
+ assert exc_info.value.error_code == "app_tenant_mismatch"
+ assert exc_info.value.status_code == 403
+
+
+def test_prepare_defers_cached_quota_status_to_authoritative_reservation(
+ sqlite_session_factory: sessionmaker[Session],
+ sqlite_session: Session,
+) -> None:
+ request = _request()
+ _persist_app(sqlite_session, request=request)
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+ model_instance, provider_model = _model_instance(status=ModelStatus.QUOTA_EXCEEDED)
+
+ prepared = _prepare(service, request, model_instance)
+
+ assert prepared.model_instance is model_instance
+ provider_model.raise_for_status.assert_not_called()
+
+
+def test_gateway_uses_quota_managed_instance_as_single_credit_owner(
+ sqlite_session_factory: sessionmaker[Session],
+ sqlite_session: Session,
+) -> None:
+ request = _request()
+ _persist_app(sqlite_session, request=request)
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+ model_instance, _ = _model_instance()
+ reservation = MagicMock(commit_before_delivery=True)
+ model_instance.reserve_quota = MagicMock(return_value=reservation)
+ prepared = _prepare(service, request, model_instance)
+ provider_chunk = _chunk("done", usage=_usage())
+
+ def provider_stream() -> Generator[LLMResultChunk, None, None]:
+ yield provider_chunk
+
+ with patch.object(ModelInstance, "invoke_llm", return_value=provider_stream()) as provider_invoke:
+ chunks = list(service.invoke(prepared))
+
+ assert chunks == [provider_chunk]
+ model_instance.reserve_quota.assert_called_once_with(request_id=request.caller.invocation_id)
+ reservation.commit.assert_called_once_with(provider_chunk.delta.usage)
+ reservation.release.assert_called_once_with()
+ provider_invoke.assert_called_once()
+
+
+def test_gateway_forwards_prompt_messages_without_revalidation() -> None:
+ request = _request()
+ prompt_messages: list[dict[str, JsonValue]] = [
+ {
+ "role": "assistant",
+ "content": "",
+ "tool_calls": [
+ {
+ "id": "call-1",
+ "type": "function",
+ "function": {"name": "search", "arguments": '{"query":"Dify"}'},
+ }
+ ],
+ "provider_extension": {"thought_signature": "opaque"},
+ },
+ {
+ "role": "tool",
+ "content": "result",
+ "name": "search",
+ "tool_call_id": "call-1",
+ },
+ ]
+ request.target.prompt_messages = prompt_messages
+ model_instance = MagicMock(spec=ModelInstance)
+ model_instance.invoke_llm.return_value = iter([])
+ prepared = PreparedAgentLLMInvocation(request=request, model_instance=model_instance)
+
+ assert list(AgentLLMInnerService().invoke(prepared)) == []
+ assert model_instance.invoke_llm.call_args.kwargs["prompt_messages"] is prompt_messages
+
+
+def test_retried_gateway_delivery_uses_one_effective_billing_charge(
+ sqlite_session_factory: sessionmaker[Session],
+ sqlite_session: Session,
+) -> None:
+ request = _request()
+ _persist_app(sqlite_session, request=request)
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+ model_instance, _ = _model_instance()
+ model_instance.provider_model_bundle.configuration.tenant_id = request.caller.tenant_id
+ prepared = _prepare(service, request, model_instance)
+ provider_configuration = SimpleNamespace(
+ using_provider_type=ProviderType.SYSTEM,
+ get_provider_model=MagicMock(return_value=SimpleNamespace(status=ModelStatus.ACTIVE)),
+ system_configuration=SimpleNamespace(
+ current_quota_type=ProviderQuotaType.TRIAL,
+ quota_configurations=[
+ SimpleNamespace(
+ quota_type=ProviderQuotaType.TRIAL,
+ quota_unit=QuotaUnit.CREDITS,
+ quota_limit=100,
+ )
+ ],
+ ),
+ )
+ reservations_by_request: dict[str, str] = {}
+ committed_reservations: set[str] = set()
+ effective_charges = 0
+ reserve_request_ids: list[str] = []
+
+ def reserve(*, request_id: str, **_: object) -> dict[str, object]:
+ reserve_request_ids.append(request_id)
+ reservation_id = reservations_by_request.setdefault(request_id, "reservation-1")
+ return {"reservation_id": reservation_id, "available": 97, "reserved": 3}
+
+ def commit(*, reservation_id: str, actual_amount: int, **_: object) -> dict[str, object]:
+ nonlocal effective_charges
+ if reservation_id not in committed_reservations:
+ committed_reservations.add(reservation_id)
+ effective_charges += actual_amount
+ return {"available": 97, "reserved": 0, "refunded": 0}
+
+ def provider_invoke(*_: object, **__: object) -> Generator[LLMResultChunk, None, None]:
+ yield _chunk("done", usage=_usage())
+
+ with (
+ patch("core.app.llm.quota._get_provider_configuration", return_value=provider_configuration),
+ patch.object(type(dify_config), "get_model_credits", return_value=3),
+ patch("services.credit_pool_service.CreditPoolService._use_billing_quota", return_value=True),
+ patch("services.billing_service.BillingService.quota_reserve", side_effect=reserve),
+ patch("services.billing_service.BillingService.quota_commit", side_effect=commit),
+ patch.object(ModelInstance, "invoke_llm", side_effect=provider_invoke),
+ ):
+ first = list(service.invoke(prepared))
+ second = list(service.invoke(prepared))
+
+ assert first[0].delta.message.content == "done"
+ assert second[0].delta.message.content == "done"
+ assert reserve_request_ids == [request.caller.invocation_id, request.caller.invocation_id]
+ assert reservations_by_request == {request.caller.invocation_id: "reservation-1"}
+ assert committed_reservations == {"reservation-1"}
+ assert effective_charges == 3
+
+
+def test_gateway_releases_reservation_when_provider_fails_before_delivery(
+ sqlite_session_factory: sessionmaker[Session],
+ sqlite_session: Session,
+) -> None:
+ request = _request()
+ _persist_app(sqlite_session, request=request)
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+ model_instance, _ = _model_instance()
+ reservation = MagicMock(commit_before_delivery=True)
+ model_instance.reserve_quota = MagicMock(return_value=reservation)
+ prepared = _prepare(service, request, model_instance)
+
+ def failing_stream() -> Generator[LLMResultChunk, None, None]:
+ raise RuntimeError("provider failed")
+ yield
+
+ with patch.object(ModelInstance, "invoke_llm", return_value=failing_stream()):
+ with pytest.raises(RuntimeError, match="provider failed"):
+ list(service.invoke(prepared))
+
+ reservation.commit.assert_not_called()
+ reservation.release.assert_called_once_with()
+
+
+def test_gateway_buffers_usage_based_quota_until_terminal_usage(
+ sqlite_session_factory: sessionmaker[Session],
+ sqlite_session: Session,
+) -> None:
+ request = _request()
+ _persist_app(sqlite_session, request=request)
+ service = AgentLLMInnerService(session_factory=sqlite_session_factory)
+ model_instance, _ = _model_instance()
+ events: list[str] = []
+ reservation = MagicMock(commit_before_delivery=False)
+ reservation.commit.side_effect = lambda _usage: events.append("commit")
+ model_instance.reserve_quota = MagicMock(return_value=reservation)
+ prepared = _prepare(service, request, model_instance)
+ terminal_usage = _usage(total_tokens=21)
+
+ def provider_stream() -> Generator[LLMResultChunk, None, None]:
+ events.append("provider:first")
+ yield _chunk("first")
+ events.append("provider:last")
+ yield _chunk("last", usage=terminal_usage)
+
+ with patch.object(ModelInstance, "invoke_llm", return_value=provider_stream()):
+ stream = service.invoke(prepared)
+ first = next(stream)
+ events.append("consumer:first")
+ rest = list(stream)
+
+ assert first.delta.message.content == "first"
+ assert [chunk.delta.message.content for chunk in rest] == ["last"]
+ assert events == ["provider:first", "provider:last", "commit", "consumer:first"]
+ reservation.commit.assert_called_once_with(terminal_usage)
+ reservation.release.assert_called_once_with()
diff --git a/api/tests/unit_tests/services/test_credit_pool_service.py b/api/tests/unit_tests/services/test_credit_pool_service.py
index 304411c8f94..2aeaf5313f0 100644
--- a/api/tests/unit_tests/services/test_credit_pool_service.py
+++ b/api/tests/unit_tests/services/test_credit_pool_service.py
@@ -394,6 +394,45 @@ def test_check_and_deduct_credits_uses_billing_reserve_and_commit_when_enabled()
quota_release.assert_not_called()
+def test_check_and_deduct_credits_forwards_deterministic_billing_identity() -> None:
+ with (
+ patch("services.credit_pool_service.dify_config.DEPLOYMENT_EDITION", DeploymentEdition.CLOUD),
+ patch("services.billing_service.BillingService.quota_reserve") as quota_reserve,
+ patch("services.billing_service.BillingService.quota_commit") as quota_commit,
+ ):
+ quota_reserve.return_value = {"reservation_id": "reservation-1", "available": 7, "reserved": 3}
+
+ result = CreditPoolService.check_and_deduct_credits(
+ tenant_id="tenant-1",
+ credits_required=3,
+ pool_type="trial",
+ request_id="invocation-1",
+ metadata={"agent_run_id": "run-1"},
+ )
+
+ assert result == 3
+ expected_metadata = {
+ "source": "credit_pool.check_and_deduct",
+ "agent_run_id": "run-1",
+ }
+ quota_reserve.assert_called_once_with(
+ tenant_id="tenant-1",
+ feature_key=FEATURE_KEY_CREDIT_POOL,
+ bucket="trial",
+ request_id="invocation-1",
+ amount=3,
+ meta=expected_metadata,
+ )
+ quota_commit.assert_called_once_with(
+ tenant_id="tenant-1",
+ feature_key=FEATURE_KEY_CREDIT_POOL,
+ bucket="trial",
+ reservation_id="reservation-1",
+ actual_amount=3,
+ meta=expected_metadata,
+ )
+
+
def test_check_and_deduct_credits_raises_when_billing_reserve_is_insufficient() -> None:
with (
patch("services.credit_pool_service.dify_config.DEPLOYMENT_EDITION", DeploymentEdition.CLOUD),
diff --git a/dify-agent/docs/dify-agent/get-started/index.md b/dify-agent/docs/dify-agent/get-started/index.md
index 71990baad5a..06bc0950015 100644
--- a/dify-agent/docs/dify-agent/get-started/index.md
+++ b/dify-agent/docs/dify-agent/get-started/index.md
@@ -144,7 +144,7 @@ uv run --extra server uvicorn dify_agent.server.app:app \
In another shell, keep working from the `dify-agent` directory. Create
`run_dify_agent_client.py` with the example below, then replace the placeholder
-tenant id and provider credential values.
+tenant, user, app, provider, and model values.
```python {test="skip" lint="skip"}
import asyncio
@@ -165,14 +165,12 @@ API_BASE_URL = "http://127.0.0.1:8000"
TENANT_ID = "replace-with-tenant-id"
PLUGIN_ID = "langgenius/openai"
-USER_ID: str | None = None
+USER_ID = "replace-with-user-id"
+APP_ID = "replace-with-app-id"
# Keep these aligned with DIFY_AGENT_PROVIDER and DIFY_AGENT_MODEL_NAME in dify-agent/.env.
MODEL_PROVIDER = "replace-with-provider-from-dify-agent-env"
MODEL_NAME = "replace-with-model-from-dify-agent-env"
-MODEL_CREDENTIALS: dict[str, str | int | float | bool | None] = {
- "api_key": "replace-with-provider-key",
-}
SYSTEM_PROMPT = "You are a concise assistant."
USER_PROMPT = "用一句话介绍 Dify Agent。"
@@ -193,7 +191,10 @@ def build_request() -> CreateRunRequest:
config=DifyExecutionContextLayerConfig(
tenant_id=TENANT_ID,
user_id=USER_ID,
- invoke_from="workflow_run",
+ user_from="account",
+ app_id=APP_ID,
+ agent_mode="single_step",
+ invoke_from="debugger",
),
),
RunLayerSpec(
@@ -204,7 +205,6 @@ def build_request() -> CreateRunRequest:
plugin_id=PLUGIN_ID,
model_provider=MODEL_PROVIDER,
model=MODEL_NAME,
- credentials=MODEL_CREDENTIALS,
),
),
],
@@ -238,9 +238,9 @@ if __name__ == "__main__":
## Run the client example
-The server-side `.env` controls how Dify Agent reaches the plugin daemon. The
-client example controls which tenant/plugin/provider/model and provider
-credentials the run uses.
+The server-side `.env` controls how Dify Agent reaches Dify API and the plugin
+daemon. The client example selects the tenant/user/app/plugin/provider/model;
+Dify API resolves model credentials at invocation time.
Run the example from the `dify-agent` directory:
@@ -248,10 +248,6 @@ Run the example from the `dify-agent` directory:
uv run python ./run_dify_agent_client.py
```
-The shape of `MODEL_CREDENTIALS` depends on the selected plugin provider's
-credential schema. The `{"api_key":"..."}` value above is only an OpenAI-style
-example.
-
Set `MODEL_PROVIDER` and `MODEL_NAME` to the same values as
`DIFY_AGENT_PROVIDER` and `DIFY_AGENT_MODEL_NAME` in `dify-agent/.env`.
@@ -261,9 +257,7 @@ If the run fails, check these items first:
1. Redis is running and reachable from the Dify Agent server.
2. The Dify Agent server is listening on `127.0.0.1:8000`.
-3. `DIFY_AGENT_PLUGIN_DAEMON_URL` points to the correct plugin daemon.
-4. `DIFY_AGENT_PLUGIN_DAEMON_API_KEY` matches the plugin daemon server key.
+3. `DIFY_AGENT_INNER_API_URL` points to the Dify API used by this environment.
+4. `DIFY_AGENT_INNER_API_KEY` matches Dify API's configured inner API key.
5. `PLUGIN_ID`, `MODEL_PROVIDER`, and `MODEL_NAME` in the client example match
- the corresponding values configured in `dify-agent/.env` and a provider
- available through the plugin daemon.
-6. `MODEL_CREDENTIALS` matches that provider's credential schema.
+ a model provider configured for the selected tenant in Dify API.
diff --git a/dify-agent/docs/dify-agent/guide/index.md b/dify-agent/docs/dify-agent/guide/index.md
index 75f5a4f2ba6..56996a5de5c 100644
--- a/dify-agent/docs/dify-agent/guide/index.md
+++ b/dify-agent/docs/dify-agent/guide/index.md
@@ -374,4 +374,4 @@ The repository includes simple consumers that print observed output/events:
consumes raw SSE frames for an existing run id.
The create-run examples submit Dify plugin model layers, so they require Redis,
-the API server, plugin-daemon settings, and provider credentials.
+the Agent server, Dify API gateway settings, and a configured model provider in Dify.
diff --git a/dify-agent/docs/dify-agent/user-manual/ask-human-layer/index.md b/dify-agent/docs/dify-agent/user-manual/ask-human-layer/index.md
index 404819398bc..fc00b58aef5 100644
--- a/dify-agent/docs/dify-agent/user-manual/ask-human-layer/index.md
+++ b/dify-agent/docs/dify-agent/user-manual/ask-human-layer/index.md
@@ -38,6 +38,10 @@ from agenton_collections.layers.plain import PromptLayerConfig
from agenton_collections.layers.pydantic_ai import PYDANTIC_AI_HISTORY_LAYER_TYPE_ID
from dify_agent.layers.ask_human import DIFY_ASK_HUMAN_LAYER_TYPE_ID, DifyAskHumanLayerConfig
from dify_agent.layers.dify_plugin import DifyPluginLLMLayerConfig
+from dify_agent.layers.execution_context import (
+ DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
+ DifyExecutionContextLayerConfig,
+)
from dify_agent.protocol import DIFY_AGENT_HISTORY_LAYER_ID, DIFY_AGENT_MODEL_LAYER_ID
from dify_agent.protocol.schemas import CreateRunRequest, RunComposition, RunLayerSpec
@@ -53,6 +57,18 @@ request = CreateRunRequest(
user="Review the deployment plan and proceed only after getting the required approval.",
),
),
+ RunLayerSpec(
+ name="execution_context",
+ type=DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
+ config=DifyExecutionContextLayerConfig(
+ tenant_id="replace-with-tenant-id",
+ user_id="replace-with-user-id",
+ user_from="account",
+ app_id="replace-with-app-id",
+ agent_mode="single_step",
+ invoke_from="debugger",
+ ),
+ ),
RunLayerSpec(
name=DIFY_AGENT_HISTORY_LAYER_ID,
type=PYDANTIC_AI_HISTORY_LAYER_TYPE_ID,
@@ -70,11 +86,11 @@ request = CreateRunRequest(
RunLayerSpec(
name=DIFY_AGENT_MODEL_LAYER_ID,
type="dify.plugin.llm",
+ deps={"execution_context": "execution_context"},
config=DifyPluginLLMLayerConfig(
plugin_id="langgenius/openai",
model_provider="openai",
model="gpt-5.2",
- credentials={"openai_api_key": "<redacted>"},
),
),
]
diff --git a/dify-agent/docs/dify-agent/user-manual/plugin-llm-layer/index.md b/dify-agent/docs/dify-agent/user-manual/plugin-llm-layer/index.md
index 889d67778b6..92c0923af8a 100644
--- a/dify-agent/docs/dify-agent/user-manual/plugin-llm-layer/index.md
+++ b/dify-agent/docs/dify-agent/user-manual/plugin-llm-layer/index.md
@@ -1,11 +1,11 @@
# Plugin LLM layer
The plugin LLM layer selects the plugin package, model provider, model name,
-provider credentials, and optional model settings for the current run. Dify
-Agent reads the model from the reserved layer name `llm`.
+and optional model settings for the current run. Dify Agent reads the model
+from the reserved layer name `llm`; Dify API resolves model credentials.
It must depend on an [execution context layer](../execution-context-layer/index.md),
-because that layer supplies the daemon identity and transport context.
+because that layer supplies the caller identity required by the API gateway.
## Config fields
@@ -14,7 +14,6 @@ because that layer supplies the daemon identity and transport context.
| `plugin_id` | `str` | Plugin package id, for example `langgenius/openai`. |
| `model_provider` | `str` | Provider name inside `plugin_id`. Use the value of `DIFY_AGENT_PROVIDER` from `dify-agent/.env`. |
| `model` | `str` | Model name. Use the value of `DIFY_AGENT_MODEL_NAME` from `dify-agent/.env`. |
-| `credentials` | `dict[str, str \| int \| float \| bool \| None]` | Provider-specific credential object. |
| `model_settings` | `ModelSettings \| None` | Optional pydantic-ai model settings. |
The plugin LLM layer type id is `dify.plugin.llm`.
@@ -38,7 +37,6 @@ llm_layer = RunLayerSpec(
plugin_id=PLUGIN_ID,
model_provider=MODEL_PROVIDER,
model=MODEL_NAME,
- credentials={"api_key": "replace-with-provider-key"},
),
)
```
@@ -80,7 +78,11 @@ composition = RunComposition(
type=DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
config=DifyExecutionContextLayerConfig(
tenant_id="replace-with-tenant-id",
- invoke_from="workflow_run",
+ user_id="replace-with-user-id",
+ user_from="account",
+ app_id="replace-with-app-id",
+ agent_mode="single_step",
+ invoke_from="debugger",
),
),
RunLayerSpec(
@@ -91,7 +93,6 @@ composition = RunComposition(
plugin_id=PLUGIN_ID,
model_provider=MODEL_PROVIDER,
model=MODEL_NAME,
- credentials={"api_key": "replace-with-provider-key"},
),
),
]
@@ -102,9 +103,6 @@ composition = RunComposition(
- The model layer must use the reserved name `llm` (`DIFY_AGENT_MODEL_LAYER_ID`).
- `plugin_id` belongs here because model calls are plugin-specific business
- calls. The shared execution-context layer only carries Dify run and
- tenant/user daemon context.
-- Credential shape depends on the selected plugin provider; the OpenAI-style
- `api_key` field above is only an example.
-- Client-submitted model credentials remain in the scheduled request memory and
- are not part of run records or session snapshots.
+ calls. The shared execution-context layer carries the Dify caller context.
+- Model credentials are never accepted from the Agent request. Dify API resolves
+ the tenant's current provider configuration and owns quota accounting.
diff --git a/dify-agent/docs/dify-agent/user-manual/shell-layer/index.md b/dify-agent/docs/dify-agent/user-manual/shell-layer/index.md
index 0169af0369d..210e59f43dd 100644
--- a/dify-agent/docs/dify-agent/user-manual/shell-layer/index.md
+++ b/dify-agent/docs/dify-agent/user-manual/shell-layer/index.md
@@ -171,6 +171,9 @@ request = CreateRunRequest(
type=DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
config=DifyExecutionContextLayerConfig(
tenant_id="92cca973-2d6f-45e0-906e-0b7eda5f2ccf",
+ user_id="replace-with-user-id",
+ user_from="account",
+ app_id="replace-with-app-id",
agent_id="8d542564-159d-4168-985c-dde8d8ff6092",
agent_config_version_id="931a4cee-4434-4c1c-8fbd-0a3c7591095d",
agent_config_version_kind="snapshot",
@@ -197,7 +200,6 @@ request = CreateRunRequest(
plugin_id="langgenius/gemini",
model_provider="google",
model="gemini-2.5-flash",
- credentials={"google_api_key": "<redacted>"},
),
),
]
diff --git a/dify-agent/examples/dify_agent/dify_agent_examples/run_pydantic_ai_agent.py b/dify-agent/examples/dify_agent/dify_agent_examples/run_pydantic_ai_agent.py
index 12eb42dc1b8..6c50ca6e2d0 100644
--- a/dify-agent/examples/dify_agent/dify_agent_examples/run_pydantic_ai_agent.py
+++ b/dify-agent/examples/dify_agent/dify_agent_examples/run_pydantic_ai_agent.py
@@ -1,10 +1,9 @@
-"""Run a Pydantic AI agent through the Dify plugin-daemon adapter.
+"""Run a Pydantic AI agent through the Dify API LLM gateway.
Prerequisites:
- Sync the server runtime dependencies first: `uv sync --project dify-agent --extra server`.
-- Start the plugin daemon from `dify-aio/dify/docker/docker-compose.middleware.yaml`.
-- Run the Dify API with `dify-aio/dify/api/.env` so the daemon can resolve tenants/plugins.
-- Fill `dify-agent/.env` with a real tenant, plugin, provider, model, and provider credentials.
+- Run the Dify API with its inner Agent LLM endpoint enabled.
+- Fill `dify-agent/.env` with a real tenant, user, app, plugin, provider, and model.
This example is meant to be run from a source checkout because
`dify_agent_examples` is not part of the published package.
@@ -17,14 +16,15 @@ Example from the repository root:
from __future__ import annotations
import asyncio
-import json
import os
from pathlib import Path
-from typing import Any
+from uuid import uuid4
+import httpx
from pydantic_ai import Agent
-from dify_agent.adapters.llm import DifyLLMAdapterModel, DifyPluginDaemonProvider
+from dify_agent.adapters.llm import DifyApiLLMProvider, DifyLLMAdapterModel
+from dify_agent.layers.execution_context import DifyExecutionContextLayerConfig
PROJECT_ROOT = Path(__file__).resolve().parents[3]
@@ -50,33 +50,35 @@ def required_env(name: str) -> str:
raise RuntimeError(f"Missing required environment variable: {name}")
-def load_credentials() -> dict[str, Any]:
- raw_credentials = required_env("DIFY_AGENT_MODEL_CREDENTIALS_JSON")
- credentials = json.loads(raw_credentials)
- if not isinstance(credentials, dict):
- raise RuntimeError("DIFY_AGENT_MODEL_CREDENTIALS_JSON must be a JSON object")
- return credentials
-
-
async def main() -> None:
load_env_file(PROJECT_ROOT / ".env")
- model = DifyLLMAdapterModel(
- required_env("DIFY_AGENT_MODEL_NAME"),
- DifyPluginDaemonProvider(
- tenant_id=required_env("DIFY_AGENT_TENANT_ID"),
+ async with httpx.AsyncClient(timeout=600, trust_env=False) as http_client:
+ provider = DifyApiLLMProvider(
plugin_id=required_env("DIFY_AGENT_PLUGIN_ID"),
- plugin_daemon_url=required_env("PLUGIN_DAEMON_URL"),
- plugin_daemon_api_key=required_env("PLUGIN_DAEMON_KEY"),
- ),
- model_provider=required_env("DIFY_AGENT_PROVIDER"),
- credentials=load_credentials(),
- )
- agent = Agent(model=model)
- async with agent.run_stream("Explain the theory of relativity") as run:
- async for piece in run.stream_output():
- print(piece, end="", flush=True)
- print(run.usage())
+ inner_api_url=required_env("DIFY_INNER_API_URL"),
+ inner_api_key=required_env("DIFY_INNER_API_KEY"),
+ execution_context=DifyExecutionContextLayerConfig(
+ tenant_id=required_env("DIFY_AGENT_TENANT_ID"),
+ user_id=required_env("DIFY_AGENT_USER_ID"),
+ user_from="account",
+ app_id=required_env("DIFY_AGENT_APP_ID"),
+ agent_mode="single_step",
+ invoke_from="debugger",
+ ),
+ agent_run_id=str(uuid4()),
+ http_client=http_client,
+ )
+ model = DifyLLMAdapterModel(
+ required_env("DIFY_AGENT_MODEL_NAME"),
+ provider,
+ model_provider=required_env("DIFY_AGENT_PROVIDER"),
+ )
+ agent = Agent(model=model)
+ async with agent.run_stream("Explain the theory of relativity") as run:
+ async for piece in run.stream_output():
+ print(piece, end="", flush=True)
+ print(run.usage())
if __name__ == "__main__":
diff --git a/dify-agent/examples/dify_agent/dify_agent_examples/run_server_consumer.py b/dify-agent/examples/dify_agent/dify_agent_examples/run_server_consumer.py
index 4ac7029fca7..25c9081c1d4 100644
--- a/dify-agent/examples/dify_agent/dify_agent_examples/run_server_consumer.py
+++ b/dify-agent/examples/dify_agent/dify_agent_examples/run_server_consumer.py
@@ -20,7 +20,6 @@ from dify_agent.client import Client
from dify_agent.layers.execution_context import DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID, DifyExecutionContextLayerConfig
from dify_agent.layers.dify_plugin import (
DIFY_PLUGIN_LLM_LAYER_TYPE_ID,
- DifyPluginCredentialValue,
DifyPluginLLMLayerConfig,
)
from dify_agent.protocol import DIFY_AGENT_MODEL_LAYER_ID, CreateRunRequest, RunComposition, RunLayerSpec
@@ -28,10 +27,11 @@ from dify_agent.protocol import DIFY_AGENT_MODEL_LAYER_ID, CreateRunRequest, Run
API_BASE_URL = "http://localhost:8000"
TENANT_ID = "replace-with-tenant-id"
+USER_ID = "replace-with-user-id"
+APP_ID = "replace-with-app-id"
PLUGIN_ID = "langgenius/openai"
PLUGIN_PROVIDER = "openai"
MODEL_NAME = "gpt-4o-mini"
-MODEL_CREDENTIALS: dict[str, DifyPluginCredentialValue] = {"api_key": "replace-with-provider-key"}
async def main() -> None:
@@ -53,7 +53,9 @@ async def main() -> None:
type=DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
config=DifyExecutionContextLayerConfig(
tenant_id=TENANT_ID,
+ user_id=USER_ID,
user_from="account",
+ app_id=APP_ID,
agent_mode="workflow_run",
invoke_from="service-api",
),
@@ -66,7 +68,6 @@ async def main() -> None:
plugin_id=PLUGIN_ID,
model_provider=PLUGIN_PROVIDER,
model=MODEL_NAME,
- credentials=MODEL_CREDENTIALS,
),
),
# Minimal plugin-tools example. API callers should pass
diff --git a/dify-agent/examples/dify_agent/dify_agent_examples/run_server_sync_client.py b/dify-agent/examples/dify_agent/dify_agent_examples/run_server_sync_client.py
index 4b2540506e6..7887329b666 100644
--- a/dify-agent/examples/dify_agent/dify_agent_examples/run_server_sync_client.py
+++ b/dify-agent/examples/dify_agent/dify_agent_examples/run_server_sync_client.py
@@ -13,7 +13,6 @@ from dify_agent.client import Client
from dify_agent.layers.execution_context import DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID, DifyExecutionContextLayerConfig
from dify_agent.layers.dify_plugin import (
DIFY_PLUGIN_LLM_LAYER_TYPE_ID,
- DifyPluginCredentialValue,
DifyPluginLLMLayerConfig,
)
from dify_agent.protocol import DIFY_AGENT_MODEL_LAYER_ID, CreateRunRequest, RunComposition, RunLayerSpec
@@ -21,10 +20,11 @@ from dify_agent.protocol import DIFY_AGENT_MODEL_LAYER_ID, CreateRunRequest, Run
API_BASE_URL = "http://localhost:8000"
TENANT_ID = "replace-with-tenant-id"
+USER_ID = "replace-with-user-id"
+APP_ID = "replace-with-app-id"
PLUGIN_ID = "langgenius/openai"
PLUGIN_PROVIDER = "openai"
MODEL_NAME = "gpt-4o-mini"
-MODEL_CREDENTIALS: dict[str, DifyPluginCredentialValue] = {"api_key": "replace-with-provider-key"}
def main() -> None:
@@ -46,7 +46,9 @@ def main() -> None:
type=DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
config=DifyExecutionContextLayerConfig(
tenant_id=TENANT_ID,
+ user_id=USER_ID,
user_from="account",
+ app_id=APP_ID,
agent_mode="workflow_run",
invoke_from="service-api",
),
@@ -59,7 +61,6 @@ def main() -> None:
plugin_id=PLUGIN_ID,
model_provider=PLUGIN_PROVIDER,
model=MODEL_NAME,
- credentials=MODEL_CREDENTIALS,
),
),
# Minimal plugin-tools example. API callers should pass
diff --git a/dify-agent/src/dify_agent/adapters/llm/__init__.py b/dify-agent/src/dify_agent/adapters/llm/__init__.py
index b63771ed616..4da4d38f408 100644
--- a/dify-agent/src/dify_agent/adapters/llm/__init__.py
+++ b/dify-agent/src/dify_agent/adapters/llm/__init__.py
@@ -1,6 +1,6 @@
-"""LLM adapters for Dify plugin-daemon integrations."""
+"""LLM adapters for Dify API integrations."""
from .model import DifyLLMAdapterModel
-from .provider import DifyPluginDaemonProvider
+from .provider import DifyApiLLMProvider
-__all__ = ["DifyLLMAdapterModel", "DifyPluginDaemonProvider"]
+__all__ = ["DifyApiLLMProvider", "DifyLLMAdapterModel"]
diff --git a/dify-agent/src/dify_agent/adapters/llm/model.py b/dify-agent/src/dify_agent/adapters/llm/model.py
index 3ffa4efd4e3..d9a17af2379 100644
--- a/dify-agent/src/dify_agent/adapters/llm/model.py
+++ b/dify-agent/src/dify_agent/adapters/llm/model.py
@@ -1,9 +1,8 @@
-"""Bridge Dify plugin-daemon LLM invocations into Pydantic AI's model interface.
+"""Bridge Dify API LLM invocations into Pydantic AI's model interface.
-The API and agent layers are clients of the plugin daemon, not direct hosts of provider SDK
-implementations. This adapter therefore targets the plugin-daemon dispatch protocol and maps
-Pydantic AI messages into the daemon's Graphon-compatible request and stream response schema.
-Pydantic AI keeps token counts only, so the adapter separately accumulates the daemon's complete
+The agent calls Dify API's trusted LLM gateway rather than hosting provider SDK implementations.
+This adapter maps Pydantic AI messages into the gateway's Graphon-compatible request and stream
+response schema. Pydantic AI keeps token counts only, so the adapter separately accumulates Dify's complete
Graphon usage for the lifetime of one model instance. The Agent runner creates one adapter per run
and reads that accumulated usage after all model/tool rounds finish.
"""
@@ -65,10 +64,11 @@ from pydantic_ai.messages import (
)
from pydantic_ai.models import Model, ModelRequestParameters, StreamedResponse
from pydantic_ai.profiles import ModelProfileSpec
+from pydantic_ai.providers import Provider
from pydantic_ai.settings import ModelSettings
from pydantic_ai.usage import RequestUsage
-from .provider import DifyPluginDaemonLLMClient, DifyPluginDaemonProvider
+from .provider import DifyLLMClient
_THINK_START = "<think>\n"
_THINK_END = "\n</think>"
@@ -80,7 +80,6 @@ _DETAIL_HIGH = "high"
@dataclass(slots=True)
class _DifyRequestInput:
- credentials: dict[str, object]
prompt_messages: list[PromptMessage]
model_parameters: dict[str, object]
tools: list[PromptMessageTool] | None
@@ -88,8 +87,8 @@ class _DifyRequestInput:
@dataclass(slots=True)
-class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
- """Use a Dify plugin-daemon transport and retain complete usage for one Agent run.
+class DifyLLMAdapterModel(Model[DifyLLMClient]):
+ """Use Dify API's LLM transport and retain complete usage for one Agent run.
A model instance belongs to one runner invocation. Pydantic AI may call it repeatedly while
resolving tools or retrying structured output; those sequential requests are accumulated so
@@ -97,10 +96,9 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
"""
model: str
- daemon_provider: DifyPluginDaemonProvider
+ dify_provider: Provider[DifyLLMClient]
_: KW_ONLY
model_provider: str
- credentials: dict[str, object] = field(default_factory=dict, repr=False)
model_profile: InitVar[ModelProfileSpec | None] = None
model_settings: InitVar[ModelSettings | None] = None
_accumulated_usage: LLMUsage | None = field(default=None, init=False, repr=False)
@@ -113,13 +111,13 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
Model.__init__(
self,
settings=model_settings,
- profile=model_profile or self.daemon_provider.model_profile(self.model),
+ profile=model_profile or self.dify_provider.model_profile(self.model),
)
@property
@override
- def provider(self) -> DifyPluginDaemonProvider:
- return self.daemon_provider
+ def provider(self) -> Provider[DifyLLMClient]:
+ return self.dify_provider
@property
@override
@@ -129,11 +127,11 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
@property
@override
def system(self) -> str:
- return self.daemon_provider.name
+ return self.dify_provider.name
@property
def accumulated_usage(self) -> LLMUsage | None:
- """Return complete daemon usage accumulated across successful model requests."""
+ """Return complete Dify usage accumulated across successful model requests."""
return self._accumulated_usage
@override
@@ -148,10 +146,9 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
response = DifyStreamedResponse(
model_request_parameters=prepared_params,
- chunks=self.daemon_provider.client.iter_llm_result_chunks(
+ chunks=self.dify_provider.client.iter_llm_result_chunks(
provider=self.model_provider,
model=self.model_name,
- credentials=request_input.credentials,
prompt_messages=request_input.prompt_messages,
model_parameters=request_input.model_parameters,
tools=request_input.tools,
@@ -182,10 +179,9 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
response = DifyStreamedResponse(
model_request_parameters=prepared_params,
- chunks=self.daemon_provider.client.iter_llm_result_chunks(
+ chunks=self.dify_provider.client.iter_llm_result_chunks(
provider=self.model_provider,
model=self.model_name,
- credentials=request_input.credentials,
prompt_messages=request_input.prompt_messages,
model_parameters=request_input.model_parameters,
tools=request_input.tools,
@@ -199,7 +195,7 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
self._record_usage(response.dify_usage)
def _record_usage(self, usage: LLMUsage | None) -> None:
- """Add one completed daemon request to this run's usage total."""
+ """Add one completed Dify request to this run's usage total."""
if usage is None:
return
if self._accumulated_usage is None:
@@ -214,7 +210,6 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
model_request_parameters: ModelRequestParameters,
) -> _DifyRequestInput:
return _DifyRequestInput(
- credentials=dict(self.credentials),
prompt_messages=_map_messages_to_prompt_messages(messages, model_request_parameters),
model_parameters=_map_model_settings_to_parameters(model_settings),
tools=_map_tool_definitions_to_prompt_tools(model_request_parameters),
@@ -224,7 +219,7 @@ class DifyLLMAdapterModel(Model[DifyPluginDaemonLLMClient]):
@dataclass
class DifyStreamedResponse(StreamedResponse):
- """Map one daemon response while retaining its latest complete usage payload.
+ """Map one Dify response while retaining its latest complete usage payload.
Some providers may repeat cumulative usage on more than one stream chunk. Keeping the latest
payload lets the owning model count each request exactly once instead of summing stream chunks.
@@ -262,7 +257,7 @@ class DifyStreamedResponse(StreamedResponse):
@property
def dify_usage(self) -> LLMUsage | None:
- """Return the daemon's complete usage for this model request."""
+ """Return Dify's complete usage for this model request."""
return self._dify_usage
@property
diff --git a/dify-agent/src/dify_agent/adapters/llm/provider.py b/dify-agent/src/dify_agent/adapters/llm/provider.py
index a210cce1e32..7c5176a232e 100644
--- a/dify-agent/src/dify_agent/adapters/llm/provider.py
+++ b/dify-agent/src/dify_agent/adapters/llm/provider.py
@@ -1,6 +1,6 @@
-"""Dify plugin-daemon provider for Pydantic AI LLM adapters.
+"""Dify API provider for Pydantic AI LLM adapters.
-The Pydantic AI provider represents daemon/plugin transport identity. Business
+The Pydantic AI provider represents API/plugin transport identity. Business
model provider names such as ``openai`` are request-level model identity and are
passed by ``DifyLLMAdapterModel`` for each invocation instead of being stored on
this provider.
@@ -8,9 +8,12 @@ this provider.
from __future__ import annotations
-from collections.abc import AsyncIterator, Callable, Mapping
+import json
+from collections.abc import AsyncIterator
from dataclasses import dataclass, field
-from typing import NoReturn
+from itertools import count
+from typing import NoReturn, Protocol, cast
+from uuid import NAMESPACE_URL, uuid5
import httpx
from graphon.model_runtime.entities.llm_entities import LLMResultChunk
@@ -21,13 +24,30 @@ from typing_extensions import override
from pydantic_ai.exceptions import ModelAPIError, ModelHTTPError, UnexpectedModelBehavior, UserError
from pydantic_ai.providers import Provider
+from dify_agent.layers.execution_context import DifyExecutionContextLayerConfig
from dify_agent.plugin_daemon_transport import (
decode_plugin_daemon_error_payload,
to_plugin_daemon_jsonable,
unwrap_plugin_daemon_error,
)
-_DEFAULT_DAEMON_TIMEOUT: float | httpx.Timeout | None = 600.0
+
+class DifyLLMClient(Protocol):
+ """Transport contract consumed by the Pydantic AI model adapter."""
+
+ http_client: httpx.AsyncClient
+
+ def iter_llm_result_chunks(
+ self,
+ *,
+ provider: str,
+ model: str,
+ prompt_messages: list[PromptMessage],
+ model_parameters: dict[str, object],
+ tools: list[PromptMessageTool] | None,
+ stop: list[str] | None,
+ stream: bool,
+ ) -> AsyncIterator[LLMResultChunk]: ...
class PluginDaemonBasicResponse(BaseModel):
@@ -37,161 +57,140 @@ class PluginDaemonBasicResponse(BaseModel):
@dataclass(slots=True)
-class DifyPluginDaemonLLMClient:
- """HTTP client wrapper for plugin-daemon LLM dispatch requests."""
+class DifyApiLLMClient:
+ """HTTP client for the API-owned Agent LLM metering gateway."""
- plugin_daemon_url: str
- plugin_daemon_api_key: str
- tenant_id: str
plugin_id: str
- user_id: str | None
+ inner_api_url: str
+ inner_api_key: str = field(repr=False)
+ execution_context: DifyExecutionContextLayerConfig
+ agent_run_id: str
http_client: httpx.AsyncClient = field(repr=False)
+ _call_counter: count[int] = field(default_factory=lambda: count(1), init=False, repr=False)
def __post_init__(self) -> None:
- self.plugin_daemon_url = self.plugin_daemon_url.rstrip("/")
+ self.inner_api_url = self.inner_api_url.rstrip("/")
async def iter_llm_result_chunks(
self,
*,
provider: str,
model: str,
- credentials: dict[str, object],
prompt_messages: list[PromptMessage],
model_parameters: dict[str, object],
tools: list[PromptMessageTool] | None,
stop: list[str] | None,
stream: bool,
) -> AsyncIterator[LLMResultChunk]:
- async for item in self._iter_stream_response(
- model_name=model,
- path=f"plugin/{self.tenant_id}/dispatch/llm/invoke",
- request_data={
- "provider": provider,
- "model_type": "llm",
- "model": model,
- "credentials": credentials,
- "prompt_messages": prompt_messages,
- "model_parameters": model_parameters,
- "tools": tools,
- "stop": stop,
- "stream": stream,
- },
- response_model=LLMResultChunk,
- ):
- yield item
-
- async def _iter_stream_response[T: BaseModel](
- self,
- *,
- model_name: str,
- path: str,
- request_data: Mapping[str, object],
- response_model: type[T],
- ) -> AsyncIterator[T]:
- payload: dict[str, object] = {"data": to_plugin_daemon_jsonable(request_data)}
- if self.user_id is not None:
- payload["user_id"] = self.user_id
-
+ call_index = next(self._call_counter)
+ invocation_id = str(uuid5(NAMESPACE_URL, f"dify-agent:{self.agent_run_id}:llm:{call_index}"))
+ context = self.execution_context
+ missing = [
+ field_name
+ for field_name, value in (
+ ("user_id", context.user_id),
+ ("user_from", context.user_from),
+ ("app_id", context.app_id),
+ )
+ if value is None
+ ]
+ if missing:
+ raise UserError(f"Agent LLM Gateway requires execution context fields: {', '.join(missing)}")
+
+ caller = context.model_dump(mode="json")
+ caller.update(
+ {
+ "invocation_id": invocation_id,
+ "agent_run_id": self.agent_run_id,
+ "call_index": call_index,
+ }
+ )
+ provider_id = provider if provider.count("/") == 2 else f"{self.plugin_id}/{provider}"
+ payload = to_plugin_daemon_jsonable(
+ {
+ "caller": caller,
+ "target": {
+ "provider": provider_id,
+ "model": model,
+ "prompt_messages": prompt_messages,
+ "model_parameters": model_parameters,
+ "tools": tools,
+ "stop": stop,
+ "stream": stream,
+ },
+ }
+ )
+ url = f"{self.inner_api_url}/inner/api/agent/llm/invoke"
headers = {
- "X-Api-Key": self.plugin_daemon_api_key,
- "X-Plugin-ID": self.plugin_id,
+ "X-Inner-Api-Key": self.inner_api_key,
"Content-Type": "application/json",
}
- url = f"{self.plugin_daemon_url}/{path}"
-
- async with self.http_client.stream("POST", url, headers=headers, json=payload) as response:
- if response.is_error:
- body = (await response.aread()).decode("utf-8", errors="replace")
- error = decode_plugin_daemon_error_payload(body)
- if error is not None:
- resolved_error = unwrap_plugin_daemon_error(
- error_type=error["error_type"],
- message=error["message"],
- )
- _raise_plugin_daemon_error(
- model_name=model_name,
- error_type=resolved_error["error_type"],
- message=resolved_error["message"],
+
+ try:
+ async with self.http_client.stream("POST", url, headers=headers, json=payload) as response:
+ if response.is_error:
+ body = (await response.aread()).decode("utf-8", errors="replace")
+ _raise_agent_gateway_http_error(
+ model_name=model,
status_code=response.status_code,
- body=resolved_error,
- )
- raise ModelHTTPError(response.status_code, model_name, body or None)
-
- async for raw_line in response.aiter_lines():
- line = raw_line.strip()
- if not line:
- continue
- if line.startswith("data:"):
- line = line[5:].strip()
-
- wrapped = PluginDaemonBasicResponse.model_validate_json(line)
- if wrapped.code != 0:
- error = decode_plugin_daemon_error_payload(wrapped.message)
- if error is not None:
- resolved_error = unwrap_plugin_daemon_error(
- error_type=error["error_type"],
- message=error["message"],
- )
- _raise_plugin_daemon_error(
- model_name=model_name,
- error_type=resolved_error["error_type"],
- message=resolved_error["message"],
- body=resolved_error,
- )
- raise ModelAPIError(
- model_name,
- f"Plugin daemon returned error code {wrapped.code}: {wrapped.message}",
+ body=body,
)
- if wrapped.data is None:
- raise UnexpectedModelBehavior("Plugin daemon returned an empty stream item")
- yield response_model.model_validate(wrapped.data)
+
+ async for raw_line in response.aiter_lines():
+ line = raw_line.strip()
+ if not line:
+ continue
+ if line.startswith("data:"):
+ line = line[5:].strip()
+ wrapped = PluginDaemonBasicResponse.model_validate_json(line)
+ if wrapped.code != 0:
+ error = decode_plugin_daemon_error_payload(wrapped.message)
+ if error is not None:
+ resolved_error = unwrap_plugin_daemon_error(
+ error_type=error["error_type"],
+ message=error["message"],
+ )
+ _raise_plugin_daemon_error(
+ model_name=model,
+ error_type=resolved_error["error_type"],
+ message=resolved_error["message"],
+ body=resolved_error,
+ )
+ raise ModelAPIError(model, wrapped.message)
+ if wrapped.data is None:
+ raise UnexpectedModelBehavior("Agent LLM Gateway returned an empty stream item")
+ yield LLMResultChunk.model_validate(wrapped.data)
+ except (httpx.InvalidURL, httpx.UnsupportedProtocol) as exc:
+ raise UserError(f"Agent LLM Gateway is misconfigured: {exc}") from exc
+ except httpx.TimeoutException as exc:
+ raise ModelHTTPError(504, model, "Agent LLM Gateway timed out") from exc
+ except httpx.RequestError as exc:
+ raise ModelHTTPError(503, model, f"Agent LLM Gateway request failed: {exc}") from exc
@dataclass(slots=True, kw_only=True)
-class DifyPluginDaemonProvider(Provider[DifyPluginDaemonLLMClient]):
- """Pydantic AI provider for Dify plugin-daemon dispatch requests.
-
- The provider ``name`` identifies the daemon/plugin context. The business LLM
- provider is supplied by each adapter model request so one daemon provider can
- serve different model-provider selections without mutating transport state.
- When ``http_client`` is omitted the provider owns an ``AsyncClient`` and the
- Pydantic AI provider context manager closes it. When an external client is
- supplied, ownership stays with the caller and provider exit leaves it open.
- """
-
- tenant_id: str
+class DifyApiLLMProvider(Provider[DifyLLMClient]):
+ """Pydantic AI provider backed by Dify API's metered model gateway."""
+
plugin_id: str
- plugin_daemon_url: str
- plugin_daemon_api_key: str = field(repr=False)
- user_id: str | None = None
- timeout: float | httpx.Timeout | None = _DEFAULT_DAEMON_TIMEOUT
- http_client: httpx.AsyncClient | None = field(default=None, repr=False)
- _client: DifyPluginDaemonLLMClient = field(init=False, repr=False)
- _own_http_client: httpx.AsyncClient | None = field(init=False, default=None, repr=False)
- _http_client_factory: Callable[[], httpx.AsyncClient] | None = field(init=False, default=None, repr=False)
+ inner_api_url: str
+ inner_api_key: str = field(repr=False)
+ execution_context: DifyExecutionContextLayerConfig
+ agent_run_id: str
+ http_client: httpx.AsyncClient = field(repr=False)
+ _client: DifyLLMClient = field(init=False, repr=False)
def __post_init__(self) -> None:
- self.plugin_daemon_url = self.plugin_daemon_url.rstrip("/")
- if self.http_client is None:
- self._http_client_factory = self._make_http_client
- http_client = self._make_http_client()
- self._own_http_client = http_client
- else:
- http_client = self.http_client
- self._own_http_client = None
- self._http_client_factory = None
- self._client = DifyPluginDaemonLLMClient(
- plugin_daemon_url=self.plugin_daemon_url,
- plugin_daemon_api_key=self.plugin_daemon_api_key,
- tenant_id=self.tenant_id,
+ self.inner_api_url = self.inner_api_url.rstrip("/")
+ self._client = DifyApiLLMClient(
plugin_id=self.plugin_id,
- user_id=self.user_id,
- http_client=http_client,
+ inner_api_url=self.inner_api_url,
+ inner_api_key=self.inner_api_key,
+ execution_context=self.execution_context,
+ agent_run_id=self.agent_run_id,
+ http_client=self.http_client,
)
- def _make_http_client(self) -> httpx.AsyncClient:
- return httpx.AsyncClient(timeout=self.timeout, trust_env=False)
-
@override
def _set_http_client(self, http_client: httpx.AsyncClient) -> None:
self._client.http_client = http_client
@@ -199,19 +198,33 @@ class DifyPluginDaemonProvider(Provider[DifyPluginDaemonLLMClient]):
@property
@override
def name(self) -> str:
- return f"DifyPlugin/{self.plugin_id}"
+ return f"DifyAPI/{self.plugin_id}"
@property
@override
def base_url(self) -> str:
- return self.plugin_daemon_url
+ return self.inner_api_url
@property
@override
- def client(self) -> DifyPluginDaemonLLMClient:
+ def client(self) -> DifyLLMClient:
return self._client
+def _raise_agent_gateway_http_error(*, model_name: str, status_code: int, body: str) -> NoReturn:
+ message = body or f"Agent LLM Gateway returned HTTP {status_code}"
+ try:
+ payload = cast(object, json.loads(body))
+ except json.JSONDecodeError:
+ payload = None
+ if isinstance(payload, dict):
+ typed_payload = cast(dict[str, object], payload)
+ candidate = typed_payload.get("message") or typed_payload.get("description")
+ if isinstance(candidate, str) and candidate:
+ message = candidate
+ raise ModelHTTPError(status_code, model_name, message)
+
+
def _raise_plugin_daemon_error(
*,
model_name: str,
@@ -240,6 +253,8 @@ def _raise_plugin_daemon_error(
raise ModelHTTPError(status_code or 404, model_name, http_error_body)
case "InvokeRateLimitError":
raise ModelHTTPError(status_code or 429, model_name, http_error_body)
+ case "AgentLLMQuotaExceededError" | "QuotaExceededError":
+ raise ModelHTTPError(status_code or 429, model_name, http_error_body)
case "PluginDaemonInternalServerError" | "PluginDaemonInnerError":
raise ModelHTTPError(status_code or 500, model_name, http_error_body)
case "InvokeConnectionError" | "InvokeServerUnavailableError":
diff --git a/dify-agent/src/dify_agent/layers/dify_plugin/configs.py b/dify-agent/src/dify_agent/layers/dify_plugin/configs.py
index 25651482a57..2ff75e14010 100644
--- a/dify-agent/src/dify_agent/layers/dify_plugin/configs.py
+++ b/dify-agent/src/dify_agent/layers/dify_plugin/configs.py
@@ -17,7 +17,7 @@ re-fetch and re-merge tool declarations at execution time.
from enum import StrEnum
from typing import ClassVar, Final, Literal, TypeAlias
-from pydantic import BaseModel, ConfigDict, Field, JsonValue, field_validator
+from pydantic import BaseModel, ConfigDict, Field, JsonValue, field_validator, model_validator
from pydantic_ai.settings import ModelSettings
from agenton.layers import LayerConfig
@@ -108,11 +108,19 @@ class DifyPluginLLMLayerConfig(LayerConfig):
plugin_id: str
model_provider: str
model: str
- credentials: dict[str, DifyPluginCredentialValue] = Field(default_factory=dict)
model_settings: ModelSettings | None = None
model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid", arbitrary_types_allowed=True)
+ @model_validator(mode="before")
+ @classmethod
+ def discard_legacy_credentials(cls, value: object) -> object:
+ """Accept old clients without retaining or forwarding their LLM credentials."""
+ if isinstance(value, dict) and "credentials" in value:
+ value = dict(value)
+ value.pop("credentials", None)
+ return value
+
class DifyPluginToolConfig(LayerConfig):
"""Public config for exposing one plugin tool to the agent model.
diff --git a/dify-agent/src/dify_agent/layers/dify_plugin/llm_layer.py b/dify-agent/src/dify_agent/layers/dify_plugin/llm_layer.py
index bdd9d947565..71e26f22b9e 100644
--- a/dify-agent/src/dify_agent/layers/dify_plugin/llm_layer.py
+++ b/dify-agent/src/dify_agent/layers/dify_plugin/llm_layer.py
@@ -1,23 +1,23 @@
"""Dify plugin LLM model layer.
This layer owns model capability resolution for Dify plugin-backed LLMs. It
-depends on ``DifyExecutionContextLayer`` for shared daemon settings through
+depends on ``DifyExecutionContextLayer`` for shared request context through
Agenton's direct dependency binding and returns a Pydantic AI model adapter
configured from the public LLM layer DTO. Runtime code supplies the FastAPI
lifespan-owned shared HTTP client to ``get_model``; the layer does not own or
-discover live resources. The daemon provider carries plugin transport identity,
+discover live resources. The API provider carries plugin transport identity,
while the DTO's ``model_provider`` is passed to the adapter as request-level
model identity.
"""
-from dataclasses import dataclass
+from dataclasses import dataclass, field
from typing import ClassVar
import httpx
from typing_extensions import Self, override
from agenton.layers import LayerDeps, PlainLayer
-from dify_agent.adapters.llm import DifyLLMAdapterModel
+from dify_agent.adapters.llm import DifyApiLLMProvider, DifyLLMAdapterModel
from dify_agent.layers.dify_plugin.configs import DIFY_PLUGIN_LLM_LAYER_TYPE_ID, DifyPluginLLMLayerConfig
from dify_agent.layers.execution_context.layer import DifyExecutionContextLayer
@@ -30,11 +30,13 @@ class DifyPluginLLMDeps(LayerDeps):
@dataclass(slots=True)
class DifyPluginLLMLayer(PlainLayer[DifyPluginLLMDeps, DifyPluginLLMLayerConfig]):
- """Layer that creates the Dify plugin-daemon Pydantic AI model."""
+ """Layer that creates the Dify API-backed Pydantic AI model."""
type_id: ClassVar[str | None] = DIFY_PLUGIN_LLM_LAYER_TYPE_ID
config: DifyPluginLLMLayerConfig
+ inner_api_url: str = "http://localhost:5001"
+ inner_api_key: str = field(default="", repr=False)
@classmethod
@override
@@ -42,17 +44,32 @@ class DifyPluginLLMLayer(PlainLayer[DifyPluginLLMDeps, DifyPluginLLMLayerConfig]
"""Create the LLM layer from validated public config."""
return cls(config=config)
- def get_model(self, *, http_client: httpx.AsyncClient) -> DifyLLMAdapterModel:
- """Return the configured model using the directly bound execution context."""
- provider = self.deps.execution_context.create_daemon_provider(
+ @classmethod
+ def from_config_with_settings(
+ cls,
+ config: DifyPluginLLMLayerConfig,
+ *,
+ inner_api_url: str,
+ inner_api_key: str,
+ ) -> Self:
+ return cls(config=config, inner_api_url=inner_api_url.rstrip("/"), inner_api_key=inner_api_key)
+
+ def get_model(self, *, http_client: httpx.AsyncClient, agent_run_id: str) -> DifyLLMAdapterModel:
+ """Return the configured model through the API-owned metered gateway."""
+ if http_client.is_closed:
+ raise RuntimeError("DifyPluginLLMLayer.get_model() requires an open Dify API HTTP client.")
+ provider = DifyApiLLMProvider(
plugin_id=self.config.plugin_id,
+ inner_api_url=self.inner_api_url,
+ inner_api_key=self.inner_api_key,
+ execution_context=self.deps.execution_context.config,
+ agent_run_id=agent_run_id,
http_client=http_client,
)
return DifyLLMAdapterModel(
model=self.config.model,
- daemon_provider=provider,
+ dify_provider=provider,
model_provider=self.config.model_provider,
- credentials=dict(self.config.credentials),
model_settings=self.config.model_settings,
)
diff --git a/dify-agent/src/dify_agent/layers/execution_context/layer.py b/dify-agent/src/dify_agent/layers/execution_context/layer.py
index 55a1920f535..2e4b726bcda 100644
--- a/dify-agent/src/dify_agent/layers/execution_context/layer.py
+++ b/dify-agent/src/dify_agent/layers/execution_context/layer.py
@@ -7,8 +7,7 @@ and API key are injected by the provider factory. The layer is intentionally
config/settings-only under Agenton's state-only core: it does not open, cache,
close, or snapshot HTTP clients, and its lifecycle hooks remain the inherited
no-op hooks. Runtime code passes the FastAPI lifespan-owned shared
-``httpx.AsyncClient`` into ``create_daemon_provider`` or ``create_tool_client``
-for each invocation.
+``httpx.AsyncClient`` into ``create_tool_client`` for each tool invocation.
"""
from dataclasses import dataclass
@@ -18,7 +17,6 @@ import httpx
from typing_extensions import Self, override
from agenton.layers import EmptyRuntimeState, NoLayerDeps, PlainLayer
-from dify_agent.adapters.llm import DifyPluginDaemonProvider
from dify_agent.layers.dify_plugin.tool_client import DifyPluginDaemonToolClient
from dify_agent.layers.execution_context.configs import (
DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID,
@@ -56,25 +54,6 @@ class DifyExecutionContextLayer(PlainLayer[NoLayerDeps, DifyExecutionContextLaye
"""Create the layer from public config plus server-only daemon settings."""
return cls(config=config, daemon_url=daemon_url, daemon_api_key=daemon_api_key)
- def create_daemon_provider(self, *, plugin_id: str, http_client: httpx.AsyncClient) -> DifyPluginDaemonProvider:
- """Return a daemon provider backed by the shared plugin daemon client.
-
- Raises:
- RuntimeError: if ``http_client`` has already been closed.
- """
- if http_client.is_closed:
- raise RuntimeError(
- "DifyExecutionContextLayer.create_daemon_provider() requires an open shared HTTP client."
- )
- return DifyPluginDaemonProvider(
- tenant_id=self.config.tenant_id,
- plugin_id=plugin_id,
- plugin_daemon_url=self.daemon_url,
- plugin_daemon_api_key=self.daemon_api_key,
- user_id=self.config.user_id,
- http_client=http_client,
- )
-
def create_tool_client(self, *, plugin_id: str, http_client: httpx.AsyncClient) -> DifyPluginDaemonToolClient:
"""Return a plugin-daemon tool client backed by the shared HTTP client.
diff --git a/dify-agent/src/dify_agent/runtime/compositor_factory.py b/dify-agent/src/dify_agent/runtime/compositor_factory.py
index 429c7fce57a..771a0ecb473 100644
--- a/dify-agent/src/dify_agent/runtime/compositor_factory.py
+++ b/dify-agent/src/dify_agent/runtime/compositor_factory.py
@@ -43,8 +43,8 @@ from dify_agent.layers.ask_human.layer import DifyAskHumanLayer
from dify_agent.layers.config.layer import DifyConfigLayer
from dify_agent.layers.dify_core_tools.configs import DifyCoreToolsLayerConfig
from dify_agent.layers.dify_core_tools.layer import DifyCoreToolsLayer
+from dify_agent.layers.dify_plugin.configs import DifyPluginLLMLayerConfig, DifyPluginToolsLayerConfig
from dify_agent.layers.dify_plugin.llm_layer import DifyPluginLLMLayer
-from dify_agent.layers.dify_plugin.configs import DifyPluginToolsLayerConfig
from dify_agent.layers.dify_plugin.tools_layer import DifyPluginToolsLayer
from dify_agent.layers.drive.layer import DifyDriveLayer
from dify_agent.layers.execution_context.configs import DifyExecutionContextLayerConfig
@@ -97,7 +97,14 @@ def create_default_layer_providers(
agent_stub_token_factory=agent_stub_token_factory,
),
),
- LayerProvider.from_layer_type(DifyPluginLLMLayer),
+ LayerProvider.from_factory(
+ layer_type=DifyPluginLLMLayer,
+ create=lambda config: DifyPluginLLMLayer.from_config_with_settings(
+ DifyPluginLLMLayerConfig.model_validate(config),
+ inner_api_url=inner_api_url,
+ inner_api_key=inner_api_key,
+ ),
+ ),
LayerProvider.from_factory(
layer_type=DifyPluginToolsLayer,
create=lambda config: DifyPluginToolsLayer.from_config_with_settings(
diff --git a/dify-agent/src/dify_agent/runtime/runner.py b/dify-agent/src/dify_agent/runtime/runner.py
index 814447871e5..1b76c561cfa 100644
--- a/dify-agent/src/dify_agent/runtime/runner.py
+++ b/dify-agent/src/dify_agent/runtime/runner.py
@@ -312,7 +312,10 @@ class AgentRunRunner:
)
ask_human_layer = get_ask_human_layer(run)
llm_layer = run.get_layer(DIFY_AGENT_MODEL_LAYER_ID, DifyPluginLLMLayer)
- model = llm_layer.get_model(http_client=self.plugin_daemon_http_client)
+ model = llm_layer.get_model(
+ http_client=self.dify_api_http_client,
+ agent_run_id=self.run_id,
+ )
tools = await _resolve_run_tools(
run,
plugin_daemon_http_client=self.plugin_daemon_http_client,
diff --git a/dify-agent/src/dify_agent/server/schemas.py b/dify-agent/src/dify_agent/server/schemas.py
index 879d1b211e2..9708fe0606f 100644
--- a/dify-agent/src/dify_agent/server/schemas.py
+++ b/dify-agent/src/dify_agent/server/schemas.py
@@ -23,9 +23,8 @@ def new_run_id() -> str:
class RunRecord(BaseModel):
"""Internal representation persisted for status reads.
- Only status metadata is persisted. Create-run requests can contain model
- credentials in layer config and must remain in scheduler memory rather than
- being written to Redis.
+ Only status metadata is persisted. Create-run requests can contain sensitive
+ layer configuration and must remain in scheduler memory rather than Redis.
"""
run_id: str
diff --git a/dify-agent/src/dify_agent/storage/redis_run_store.py b/dify-agent/src/dify_agent/storage/redis_run_store.py
index 6bba2368b16..f488749c0ff 100644
--- a/dify-agent/src/dify_agent/storage/redis_run_store.py
+++ b/dify-agent/src/dify_agent/storage/redis_run_store.py
@@ -5,8 +5,8 @@ streams. HTTP event cursors are Redis stream ids; ``0-0`` means replay from the
beginning for polling and SSE. Records and streams share one retention window
that is refreshed when status or event data is written. Execution is scheduled
in-process by ``dify_agent.runtime.run_scheduler``; Redis is not a job queue, and
-create-run payloads are never persisted because layer config may include model
-credentials.
+create-run payloads are never persisted because layer config may include
+sensitive runtime configuration.
"""
from collections.abc import AsyncIterator, Awaitable
diff --git a/dify-agent/tests/local/dify_agent/adapters/llm/test_api_provider.py b/dify-agent/tests/local/dify_agent/adapters/llm/test_api_provider.py
new file mode 100644
index 00000000000..70e39610d5f
--- /dev/null
+++ b/dify-agent/tests/local/dify_agent/adapters/llm/test_api_provider.py
@@ -0,0 +1,159 @@
+import asyncio
+import json
+from uuid import NAMESPACE_URL, uuid5
+
+import httpx
+import pytest
+from graphon.model_runtime.entities.message_entities import UserPromptMessage
+from pydantic_ai.exceptions import ModelHTTPError
+
+from dify_agent.adapters.llm.provider import DifyApiLLMClient
+from dify_agent.layers.execution_context import DifyExecutionContextLayerConfig
+
+from ._test_support import build_stream_error, build_stream_response, single_text_chunk
+
+
+def _execution_context() -> DifyExecutionContextLayerConfig:
+ return DifyExecutionContextLayerConfig(
+ tenant_id="tenant-1",
+ user_id="user-1",
+ user_from="account",
+ app_id="app-1",
+ workflow_id="workflow-1",
+ workflow_run_id="workflow-run-1",
+ node_id="node-1",
+ node_execution_id="execution-1",
+ agent_config_version_kind="draft",
+ agent_mode="workflow_run",
+ invoke_from="debugger",
+ trace_id="trace-1",
+ )
+
+
+def test_api_client_uses_stable_per_run_call_identity() -> None:
+ requests: list[httpx.Request] = []
+
+ async def handler(request: httpx.Request) -> httpx.Response:
+ requests.append(request)
+ return build_stream_response(*single_text_chunk("done"))
+
+ async def scenario() -> None:
+ async with httpx.AsyncClient(transport=httpx.MockTransport(handler), trust_env=False) as http_client:
+ client = DifyApiLLMClient(
+ plugin_id="acme/custom-model",
+ inner_api_url="http://dify-api/",
+ inner_api_key="inner-secret",
+ execution_context=_execution_context(),
+ agent_run_id=run_id,
+ http_client=http_client,
+ )
+
+ for _ in range(2):
+ chunks = [
+ chunk
+ async for chunk in client.iter_llm_result_chunks(
+ provider="openai",
+ model="gpt-test",
+ prompt_messages=[UserPromptMessage(content="hello")],
+ model_parameters={"temperature": 0.2},
+ tools=None,
+ stop=None,
+ stream=True,
+ )
+ ]
+ assert chunks[0].delta.message.content == "done"
+
+ run_id = "00000000-0000-0000-0000-000000000001"
+ asyncio.run(scenario())
+
+ assert len(requests) == 2
+ first_payload = json.loads(requests[0].content)
+ second_payload = json.loads(requests[1].content)
+ assert requests[0].url == "http://dify-api/inner/api/agent/llm/invoke"
+ assert requests[0].headers["X-Inner-Api-Key"] == "inner-secret"
+ assert first_payload["caller"]["call_index"] == 1
+ assert first_payload["caller"]["invocation_id"] == str(uuid5(NAMESPACE_URL, f"dify-agent:{run_id}:llm:1"))
+ assert second_payload["caller"]["call_index"] == 2
+ assert second_payload["caller"]["invocation_id"] == str(uuid5(NAMESPACE_URL, f"dify-agent:{run_id}:llm:2"))
+ assert first_payload["caller"]["agent_config_version_kind"] == "draft"
+ assert first_payload["target"]["provider"] == "acme/custom-model/openai"
+ assert "credentials" not in first_payload["target"]
+
+
+def test_api_client_propagates_gateway_quota_error() -> None:
+ async def handler(_request: httpx.Request) -> httpx.Response:
+ return httpx.Response(
+ 429,
+ json={
+ "code": "agent_llm_quota_exceeded",
+ "message": "Insufficient Message Credits.",
+ "status": 429,
+ },
+ )
+
+ async def scenario() -> None:
+ async with httpx.AsyncClient(transport=httpx.MockTransport(handler), trust_env=False) as http_client:
+ client = DifyApiLLMClient(
+ plugin_id="langgenius/openai",
+ inner_api_url="http://dify-api",
+ inner_api_key="inner-secret",
+ execution_context=_execution_context(),
+ agent_run_id="00000000-0000-0000-0000-000000000001",
+ http_client=http_client,
+ )
+
+ with pytest.raises(ModelHTTPError) as exc_info:
+ _ = [
+ chunk
+ async for chunk in client.iter_llm_result_chunks(
+ provider="openai",
+ model="gpt-test",
+ prompt_messages=[UserPromptMessage(content="hello")],
+ model_parameters={},
+ tools=None,
+ stop=None,
+ stream=True,
+ )
+ ]
+ assert exc_info.value.status_code == 429
+ assert "Insufficient Message Credits" in str(exc_info.value)
+
+ asyncio.run(scenario())
+
+
+def test_api_client_maps_stream_quota_error_to_http_429() -> None:
+ async def handler(_request: httpx.Request) -> httpx.Response:
+ return build_stream_error(
+ "AgentLLMQuotaExceededError",
+ "Insufficient hosted model quota remaining.",
+ code=-429,
+ )
+
+ async def scenario() -> None:
+ async with httpx.AsyncClient(transport=httpx.MockTransport(handler), trust_env=False) as http_client:
+ client = DifyApiLLMClient(
+ plugin_id="langgenius/openai",
+ inner_api_url="http://dify-api",
+ inner_api_key="inner-secret",
+ execution_context=_execution_context(),
+ agent_run_id="00000000-0000-0000-0000-000000000001",
+ http_client=http_client,
+ )
+
+ with pytest.raises(ModelHTTPError) as exc_info:
+ _ = [
+ chunk
+ async for chunk in client.iter_llm_result_chunks(
+ provider="openai",
+ model="gpt-test",
+ prompt_messages=[UserPromptMessage(content="hello")],
+ model_parameters={},
+ tools=None,
+ stop=None,
+ stream=True,
+ )
+ ]
+ assert exc_info.value.status_code == 429
+ assert "Insufficient hosted model quota" in str(exc_info.value)
+
+ asyncio.run(scenario())
diff --git a/dify-agent/tests/local/dify_agent/adapters/llm/test_model.py b/dify-agent/tests/local/dify_agent/adapters/llm/test_model.py
index 163f47af8ff..4fe17b399b0 100644
--- a/dify-agent/tests/local/dify_agent/adapters/llm/test_model.py
+++ b/dify-agent/tests/local/dify_agent/adapters/llm/test_model.py
@@ -23,7 +23,8 @@ from pydantic_ai.messages import (
from pydantic_ai.models import ModelRequestParameters
from pydantic_ai.tools import ToolDefinition
-from dify_agent.adapters.llm import DifyLLMAdapterModel, DifyPluginDaemonProvider
+from dify_agent.adapters.llm import DifyApiLLMProvider, DifyLLMAdapterModel
+from dify_agent.layers.execution_context import DifyExecutionContextLayerConfig
from ._test_support import (
AssistantPromptMessage,
@@ -38,23 +39,42 @@ from ._test_support import (
class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
+ _http_clients: list[httpx.AsyncClient] = []
+
+ async def asyncSetUp(self) -> None:
+ self._http_clients: list[httpx.AsyncClient] = []
+
+ async def asyncTearDown(self) -> None:
+ for client in self._http_clients:
+ await client.aclose()
+
def make_provider(
self,
*,
- user_id: str | None = None,
+ user_id: str = "user-123",
http_client: httpx.AsyncClient | None = None,
- ) -> DifyPluginDaemonProvider:
- return DifyPluginDaemonProvider(
- tenant_id="tenant-1",
+ ) -> DifyApiLLMProvider:
+ if http_client is None:
+ http_client = httpx.AsyncClient(trust_env=False)
+ self._http_clients.append(http_client)
+ return DifyApiLLMProvider(
plugin_id="langgenius/openai",
- plugin_daemon_url="http://plugin-daemon",
- plugin_daemon_api_key="daemon-secret",
- user_id=user_id,
+ inner_api_url="http://dify-api",
+ inner_api_key="inner-secret",
+ execution_context=DifyExecutionContextLayerConfig(
+ tenant_id="tenant-1",
+ user_id=user_id,
+ user_from="account",
+ app_id="app-1",
+ agent_mode="single_step",
+ invoke_from="debugger",
+ ),
+ agent_run_id="run-1",
http_client=http_client,
)
@asynccontextmanager
- async def mock_daemon_stream(self, handler: httpx.MockTransport):
+ async def mock_gateway_stream(self, handler: httpx.MockTransport):
@asynccontextmanager
async def mock_stream(
client: httpx.AsyncClient,
@@ -73,7 +93,7 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
with patch.object(httpx.AsyncClient, "stream", new=mock_stream):
yield
- async def test_request_uses_plugin_daemon_dispatch_contract(self) -> None:
+ async def test_request_uses_api_gateway_contract(self) -> None:
messages = [
ModelRequest(
parts=[
@@ -128,17 +148,15 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(request: httpx.Request) -> httpx.Response:
self.assertEqual(request.method, "POST")
- self.assertEqual(request.url.path, "/plugin/tenant-1/dispatch/llm/invoke")
- self.assertEqual(request.headers["X-Api-Key"], "daemon-secret")
- self.assertEqual(request.headers["X-Plugin-ID"], "langgenius/openai")
+ self.assertEqual(request.url.path, "/inner/api/agent/llm/invoke")
+ self.assertEqual(request.headers["X-Inner-Api-Key"], "inner-secret")
payload = json.loads(request.content.decode("utf-8"))
- self.assertEqual(payload["user_id"], "user-123")
- data = payload["data"]
- self.assertEqual(data["provider"], "openai")
- self.assertEqual(data["model_type"], "llm")
+ self.assertEqual(payload["caller"]["user_id"], "user-123")
+ data = payload["target"]
+ self.assertEqual(data["provider"], "langgenius/openai/openai")
self.assertEqual(data["model"], "demo-model")
- self.assertEqual(data["credentials"], {"api_key": "secret"})
+ self.assertNotIn("credentials", data)
self.assertEqual(
data["model_parameters"],
{"temperature": 0.2, "max_tokens": 128, "logit_bias": {"1": 2}},
@@ -166,12 +184,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
)
)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(user_id="user-123"),
model_provider="openai",
- credentials={"api_key": "secret"},
model_settings={"temperature": 0.2, "stop_sequences": ["DEFAULT_STOP"]},
)
@@ -182,13 +199,13 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
)
self.assertEqual(response.model_name, "demo-model")
- self.assertEqual(response.provider_name, "DifyPlugin/langgenius/openai")
+ self.assertEqual(response.provider_name, "DifyAPI/langgenius/openai")
self.assertEqual(response.usage.input_tokens, 11)
self.assertEqual(response.usage.output_tokens, 7)
self.assertEqual(response.parts[0].part_kind, "text")
self.assertEqual(cast(TextPart, response.parts[0]).content, "adapter response")
- async def test_request_accumulates_complete_plugin_usage_across_model_rounds(self) -> None:
+ async def test_request_accumulates_complete_dify_usage_across_model_rounds(self) -> None:
usages = [
make_usage(
prompt_tokens=10,
@@ -236,12 +253,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
)
)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
_ = await adapter.request(
[ModelRequest(parts=[UserPromptPart("first")])],
@@ -296,7 +312,7 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(request: httpx.Request) -> httpx.Response:
payload = json.loads(request.content.decode("utf-8"))
- prompt_messages = payload["data"]["prompt_messages"]
+ prompt_messages = payload["target"]["prompt_messages"]
self.assertEqual([message["role"] for message in prompt_messages], ["system", "user", "assistant", "tool"])
self.assertEqual(prompt_messages[2]["content"], "")
@@ -308,12 +324,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
return build_stream_response(*single_text_chunk("adapter response", prompt_tokens=11, completion_tokens=7))
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
response = await adapter.request(
@@ -339,7 +354,7 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(request: httpx.Request) -> httpx.Response:
payload = json.loads(request.content.decode("utf-8"))
- prompt_messages = payload["data"]["prompt_messages"]
+ prompt_messages = payload["target"]["prompt_messages"]
tool_calls = prompt_messages[1]["tool_calls"]
self.assertEqual(tool_calls[0]["id"], "tool-call-0-lookup")
@@ -347,12 +362,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
return build_stream_response(*single_text_chunk("adapter response", prompt_tokens=11, completion_tokens=7))
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
response = await adapter.request(
@@ -379,19 +393,18 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(request: httpx.Request) -> httpx.Response:
payload = json.loads(request.content.decode("utf-8"))
- prompt_messages = payload["data"]["prompt_messages"]
+ prompt_messages = payload["target"]["prompt_messages"]
self.assertEqual([message["role"] for message in prompt_messages], ["user", "assistant", "user"])
self.assertEqual(prompt_messages[1]["content"], "<think>\nplan\n</think>answer")
return build_stream_response(*single_text_chunk("adapter response", prompt_tokens=11, completion_tokens=7))
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
response = await adapter.request(
@@ -413,19 +426,18 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(request: httpx.Request) -> httpx.Response:
payload = json.loads(request.content.decode("utf-8"))
- prompt_messages = payload["data"]["prompt_messages"]
+ prompt_messages = payload["target"]["prompt_messages"]
self.assertEqual([message["role"] for message in prompt_messages], ["system", "user", "user"])
self.assertEqual(prompt_messages[2]["content"], "follow up")
return build_stream_response(*single_text_chunk("adapter response", prompt_tokens=11, completion_tokens=7))
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
response = await adapter.request(
@@ -439,10 +451,10 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
self.assertEqual(cast(TextPart, response.parts[0]).content, "adapter response")
async def test_provider_does_not_close_external_http_client(self) -> None:
- http_client = httpx.AsyncClient()
+ http_client = httpx.AsyncClient(trust_env=False)
provider = self.make_provider(http_client=http_client)
- self.assertEqual(provider.name, "DifyPlugin/langgenius/openai")
+ self.assertEqual(provider.name, "DifyAPI/langgenius/openai")
self.assertIs(provider.client.http_client, http_client)
async with provider:
pass
@@ -454,12 +466,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(_request: httpx.Request) -> httpx.Response:
return build_stream_response(*single_text_chunk("adapter response", prompt_tokens=11, completion_tokens=7))
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
response = await adapter.request(
@@ -508,12 +519,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
),
)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
async with adapter.request_stream(
@@ -570,12 +580,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
),
)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
async with adapter.request_stream(
@@ -640,12 +649,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
),
)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
async with adapter.request_stream(
@@ -667,12 +675,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(_request: httpx.Request) -> httpx.Response:
return build_stream_response(*single_text_chunk("before<think>reasoning</think>after"))
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
response = await adapter.request(
@@ -695,12 +702,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
json.dumps({"error_type": "InvokeRateLimitError", "message": "too many"}),
)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
with self.assertRaises(ModelHTTPError) as context:
@@ -720,12 +726,11 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
def handler(_request: httpx.Request) -> httpx.Response:
return build_error_response("PluginDaemonUnauthorizedError", "invalid api key", status_code=401)
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
with self.assertRaises(ModelHTTPError) as context:
@@ -736,24 +741,17 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
)
self.assertEqual(context.exception.status_code, 401)
- self.assertEqual(
- context.exception.body,
- {
- "error_type": "PluginDaemonUnauthorizedError",
- "message": "invalid api key",
- },
- )
+ self.assertEqual(context.exception.body, "invalid api key")
async def test_request_maps_endpoint_setup_error_to_user_error(self) -> None:
def handler(_request: httpx.Request) -> httpx.Response:
return build_stream_error("EndpointSetupFailedError", "missing endpoint config")
- async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+ async with self.mock_gateway_stream(httpx.MockTransport(handler)):
adapter = DifyLLMAdapterModel(
"demo-model",
self.make_provider(),
model_provider="openai",
- credentials={"api_key": "secret"},
)
with self.assertRaises(UserError) as context:
diff --git a/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_configs.py b/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_configs.py
index a3db61a06c4..898249702a4 100644
--- a/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_configs.py
+++ b/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_configs.py
@@ -5,7 +5,6 @@ from pydantic import ValidationError
import dify_agent.layers.dify_plugin as dify_plugin_exports
from dify_agent.layers.dify_plugin import (
- DifyPluginCredentialValue,
DifyPluginLLMLayerConfig,
DifyPluginToolCredentialType,
DifyPluginToolConfig,
@@ -37,29 +36,22 @@ def test_dify_plugin_package_exports_client_safe_config_symbols_only() -> None:
assert not hasattr(dify_plugin_exports, "DifyPluginLLMLayer")
-def test_dify_plugin_llm_config_accepts_scalar_credentials_and_model_settings() -> None:
- credential: DifyPluginCredentialValue = "secret"
- config = DifyPluginLLMLayerConfig(
- plugin_id="langgenius/openai",
- model_provider="openai",
- model="gpt-4o-mini",
- credentials={"api_key": credential, "enabled": True, "retries": 2, "ratio": 0.5, "empty": None},
- model_settings={"temperature": 0.2, "max_tokens": 64},
+def test_dify_plugin_llm_config_discards_legacy_credentials() -> None:
+ config = DifyPluginLLMLayerConfig.model_validate(
+ {
+ "plugin_id": "langgenius/openai",
+ "model_provider": "openai",
+ "model": "gpt-4o-mini",
+ "credentials": {"api_key": "secret", "nested": {"legacy": True}},
+ "model_settings": {"temperature": 0.2, "max_tokens": 64},
+ }
)
assert config.plugin_id == "langgenius/openai"
assert config.model_provider == "openai"
- assert config.credentials == {"api_key": "secret", "enabled": True, "retries": 2, "ratio": 0.5, "empty": None}
+ assert not hasattr(config, "credentials")
+ assert "credentials" not in config.model_dump(mode="json")
assert config.model_settings == {"temperature": 0.2, "max_tokens": 64}
- with pytest.raises(ValidationError):
- _ = DifyPluginLLMLayerConfig.model_validate(
- {
- "plugin_id": "langgenius/openai",
- "model_provider": "openai",
- "model": "gpt-4o-mini",
- "credentials": {"nested": {"not": "allowed"}},
- }
- )
def test_dify_plugin_llm_config_rejects_old_provider_field() -> None:
diff --git a/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_layers.py b/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_layers.py
index 995fea0120c..abe020cc792 100644
--- a/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_layers.py
+++ b/dify-agent/tests/local/dify_agent/layers/dify_plugin/test_layers.py
@@ -40,7 +40,6 @@ def _llm_config() -> DifyPluginLLMLayerConfig:
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
model_settings={"temperature": 0.2},
)
@@ -54,7 +53,6 @@ def _tools_config() -> DifyPluginToolsLayerConfig:
tool_name="web_search",
credential_type="api-key",
description="Search the web.",
- credentials={"api_key": "secret"},
runtime_parameters={"api_version": "2026-01", "auth_scope": "workspace"},
parameters=_prepared_tool_parameters(),
parameters_json_schema=_prepared_tool_schema(),
@@ -72,7 +70,6 @@ def _missing_hidden_parameter_tools_config() -> DifyPluginToolsLayerConfig:
tool_name="web_search",
credential_type="api-key",
description="Search the web.",
- credentials={"api_key": "secret"},
runtime_parameters={"api_version": "2026-01"},
parameters=_prepared_tool_parameters(),
parameters_json_schema=_prepared_tool_schema(),
@@ -297,14 +294,14 @@ def test_dify_plugin_llm_layer_builds_adapter_model_from_direct_dependency() ->
execution_context = run.get_layer("renamed-execution-context", DifyExecutionContextLayer)
llm = run.get_layer("llm", DifyPluginLLMLayer)
- model = llm.get_model(http_client=client)
+ model = llm.get_model(http_client=client, agent_run_id="00000000-0000-0000-0000-000000000001")
assert llm.deps.execution_context is execution_context
assert isinstance(model, DifyLLMAdapterModel)
assert model.model_name == "demo-model"
assert model.model_provider == "openai"
- assert model.credentials == {"api_key": "secret"}
- assert model.provider.name == "DifyPlugin/langgenius/openai"
+ assert not hasattr(model, "credentials")
+ assert model.provider.name == "DifyAPI/langgenius/openai"
assert model.provider.client.http_client is client
asyncio.run(scenario())
diff --git a/dify-agent/tests/local/dify_agent/layers/execution_context/test_layer.py b/dify-agent/tests/local/dify_agent/layers/execution_context/test_layer.py
index ce093744179..cc8ccb29c78 100644
--- a/dify-agent/tests/local/dify_agent/layers/execution_context/test_layer.py
+++ b/dify-agent/tests/local/dify_agent/layers/execution_context/test_layer.py
@@ -25,25 +25,6 @@ def test_execution_context_type_id_constant_matches_implementation_class() -> No
assert DIFY_EXECUTION_CONTEXT_LAYER_TYPE_ID == DifyExecutionContextLayer.type_id
-def test_execution_context_layer_creates_daemon_provider_from_shared_http_client() -> None:
- async def scenario() -> None:
- execution_context = _execution_context_layer()
- async with httpx.AsyncClient(transport=httpx.MockTransport(lambda _request: httpx.Response(200))) as client:
- provider = execution_context.create_daemon_provider(plugin_id="langgenius/openai", http_client=client)
-
- assert provider.name == "DifyPlugin/langgenius/openai"
- assert provider.client.http_client is client
- assert provider.client.tenant_id == "tenant-1"
- assert provider.client.plugin_id == "langgenius/openai"
- assert provider.client.user_id == "user-1"
-
- async with provider:
- pass
- assert client.is_closed is False
-
- asyncio.run(scenario())
-
-
def test_execution_context_layer_creates_tool_client_from_shared_http_client() -> None:
async def scenario() -> None:
execution_context = _execution_context_layer()
@@ -64,11 +45,9 @@ def test_execution_context_layer_creates_tool_client_from_shared_http_client() -
def test_execution_context_layer_rejects_closed_shared_http_client() -> None:
async def scenario() -> None:
execution_context = _execution_context_layer()
- client = httpx.AsyncClient()
+ client = httpx.AsyncClient(trust_env=False)
await client.aclose()
- with pytest.raises(RuntimeError, match="open shared HTTP client"):
- _ = execution_context.create_daemon_provider(plugin_id="langgenius/openai", http_client=client)
with pytest.raises(RuntimeError, match="open shared HTTP client"):
_ = execution_context.create_tool_client(plugin_id="langgenius/tools", http_client=client)
@@ -102,14 +81,14 @@ def test_execution_context_layer_lifecycle_does_not_manage_http_client() -> None
}
) as run:
execution_context = run.get_layer("execution_context", DifyExecutionContextLayer)
- daemon_provider = execution_context.create_daemon_provider(
- plugin_id="langgenius/openai",
+ tool_client = execution_context.create_tool_client(
+ plugin_id="langgenius/tools",
http_client=client,
)
run.suspend_layer_on_exit("execution_context")
assert run.session_snapshot is not None
- assert daemon_provider.client.http_client is client
+ assert tool_client.http_client is client
assert client.is_closed is False
asyncio.run(scenario())
diff --git a/dify-agent/tests/local/dify_agent/protocol/test_protocol_schemas.py b/dify-agent/tests/local/dify_agent/protocol/test_protocol_schemas.py
index 5bc3915ffad..8eac3a6023e 100644
--- a/dify-agent/tests/local/dify_agent/protocol/test_protocol_schemas.py
+++ b/dify-agent/tests/local/dify_agent/protocol/test_protocol_schemas.py
@@ -160,7 +160,6 @@ def test_create_run_request_accepts_dto_first_public_composition_and_normalizes_
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
)
output_config = DifyOutputLayerConfig(
json_schema={
diff --git a/dify-agent/tests/local/dify_agent/runtime/test_run_scheduler.py b/dify-agent/tests/local/dify_agent/runtime/test_run_scheduler.py
index a22ce101d2b..4baeaa7a9b0 100644
--- a/dify-agent/tests/local/dify_agent/runtime/test_run_scheduler.py
+++ b/dify-agent/tests/local/dify_agent/runtime/test_run_scheduler.py
@@ -57,7 +57,6 @@ def _request(
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
]
diff --git a/dify-agent/tests/local/dify_agent/runtime/test_runner.py b/dify-agent/tests/local/dify_agent/runtime/test_runner.py
index e2fd7a4099c..e5ce9e1bfca 100644
--- a/dify-agent/tests/local/dify_agent/runtime/test_runner.py
+++ b/dify-agent/tests/local/dify_agent/runtime/test_runner.py
@@ -311,7 +311,6 @@ def _request(
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
]
@@ -485,7 +484,7 @@ class FakeAgentRunResult:
def test_runner_emits_terminal_success_and_snapshot(monkeypatch: pytest.MonkeyPatch) -> None:
seen_clients: list[httpx.AsyncClient] = []
- def fake_get_model(self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert self.config.model == "demo-model"
assert self.config.plugin_id == "langgenius/openai"
seen_clients.append(http_client)
@@ -558,7 +557,7 @@ def test_runner_emits_complete_plugin_usage_in_terminal_event(monkeypatch: pytes
def accumulated_usage(self) -> LLMUsage:
return complete_usage
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return PricedTestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -592,7 +591,7 @@ def test_runner_emits_complete_plugin_usage_in_terminal_event(monkeypatch: pytes
def test_runner_preserves_explicit_json_null_output(monkeypatch: pytest.MonkeyPatch) -> None:
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -625,7 +624,7 @@ def test_runner_preserves_explicit_json_null_output(monkeypatch: pytest.MonkeyPa
def test_runner_passes_explicit_step_limit_to_agent(monkeypatch: pytest.MonkeyPatch) -> None:
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -666,7 +665,7 @@ def test_runner_emits_deferred_tool_call_and_persists_pending_history(monkeypatc
tool_call_id="tool-call-1",
)
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -743,7 +742,7 @@ def test_runner_resumes_with_deferred_tool_results_and_no_user_prompt(monkeypatc
tool_call_id="tool-call-1",
)
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -848,7 +847,7 @@ def test_runner_can_emit_second_deferred_tool_call_after_resume(monkeypatch: pyt
tool_call_id="tool-call-2",
)
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -947,7 +946,7 @@ def test_runner_can_emit_second_deferred_tool_call_after_resume(monkeypatch: pyt
def test_runner_rejects_deferred_tool_call_without_history_layer(monkeypatch: pytest.MonkeyPatch) -> None:
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -991,7 +990,7 @@ def test_runner_rejects_resume_with_deferred_tool_results_without_history_layer(
) -> None:
agent_run_called = False
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -1038,7 +1037,7 @@ def test_runner_rejects_resume_with_deferred_tool_results_without_history_layer(
def test_runner_rejects_multiple_deferred_tool_calls(monkeypatch: pytest.MonkeyPatch) -> None:
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -1076,7 +1075,7 @@ def test_runner_rejects_multiple_deferred_tool_calls(monkeypatch: pytest.MonkeyP
def test_runner_rejects_deferred_approval_requests(monkeypatch: pytest.MonkeyPatch) -> None:
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="unused") # pyright: ignore[reportReturnType]
@@ -1120,7 +1119,7 @@ def test_runner_passes_dynamic_dify_plugin_tools_to_agent(monkeypatch: pytest.Mo
async def plugin_tool() -> str:
return "tool"
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -1180,7 +1179,6 @@ def test_runner_passes_dynamic_dify_plugin_tools_to_agent(monkeypatch: pytest.Mo
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -1229,7 +1227,7 @@ def test_runner_passes_dynamic_dify_knowledge_tools_to_agent(monkeypatch: pytest
async def knowledge_tool() -> str:
return "knowledge"
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -1285,7 +1283,6 @@ def test_runner_passes_dynamic_dify_knowledge_tools_to_agent(monkeypatch: pytest
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -1335,7 +1332,7 @@ def test_runner_passes_dynamic_dify_core_tools_to_agent(monkeypatch: pytest.Monk
async def core_tool() -> str:
return "core"
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -1392,7 +1389,6 @@ def test_runner_passes_dynamic_dify_core_tools_to_agent(monkeypatch: pytest.Monk
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -1442,7 +1438,7 @@ def test_runner_rejects_duplicate_tool_names_across_dynamic_tool_layers(
async def duplicate_tool() -> str:
return "tool"
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -1492,7 +1488,6 @@ def test_runner_rejects_duplicate_tool_names_across_dynamic_tool_layers(
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -1566,7 +1561,7 @@ def test_runner_rejects_duplicate_tool_names_between_static_and_dynamic_tools(
async def dynamic_duplicate_tool() -> str:
return "tool"
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -1623,7 +1618,6 @@ def test_runner_rejects_duplicate_tool_names_between_static_and_dynamic_tools(
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -1676,7 +1670,7 @@ def test_runner_rejects_duplicate_tool_names_between_shell_and_other_layers(
create_agent_called = False
shell_client = FakeRunnerShellctlClient()
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -1749,7 +1743,6 @@ def test_runner_rejects_duplicate_tool_names_between_shell_and_other_layers(
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -1801,7 +1794,7 @@ def test_runner_rejects_duplicate_tool_names_between_shell_and_other_layers(
def test_runner_passes_temporary_system_prompt_prefix_without_history_layer(monkeypatch: pytest.MonkeyPatch) -> None:
model = RecordingTestModel(custom_output_text="done")
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -1843,7 +1836,7 @@ def test_runner_prepends_current_system_prompt_to_stored_history_and_appends_onl
ModelResponse(parts=[TextPart(content="old assistant")]),
]
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -1894,7 +1887,7 @@ def test_runner_with_empty_history_layer_still_sends_system_prompt_and_saves_onl
) -> None:
model = RecordingTestModel(custom_output_text="done")
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -1944,7 +1937,7 @@ def test_runner_failure_with_history_layer_emits_failed_terminal_event_without_s
ModelResponse(parts=[TextPart(content="old assistant")]),
]
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -2004,7 +1997,7 @@ def test_runner_persists_usage_limit_failure_type_in_event_and_status(
def test_runner_applies_on_exit_overrides_to_success_snapshot(monkeypatch: pytest.MonkeyPatch) -> None:
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return TestModel(custom_output_text="done") # pyright: ignore[reportReturnType]
@@ -2070,7 +2063,7 @@ def test_runner_passes_output_layer_spec_to_agent_and_serializes_structured_resu
}
)
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -2165,7 +2158,7 @@ def test_runner_retries_invalid_structured_output_and_eventually_succeeds(monkey
]
)
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -2218,7 +2211,7 @@ def test_runner_fails_when_invalid_structured_output_exhausts_retries(monkeypatc
}
)
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
assert http_client.is_closed is False
return model # pyright: ignore[reportReturnType]
@@ -2263,7 +2256,7 @@ def test_runner_fails_when_invalid_structured_output_exhausts_retries(monkeypatc
def test_runner_rejects_invalid_output_layer_before_model_resolution(monkeypatch: pytest.MonkeyPatch) -> None:
model_requested = False
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
del http_client
nonlocal model_requested
model_requested = True
@@ -2323,7 +2316,6 @@ def test_runner_rejects_misnamed_output_layer_before_model_resolution(monkeypatc
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -2344,7 +2336,7 @@ def test_runner_rejects_misnamed_output_layer_before_model_resolution(monkeypatc
)
sink = InMemoryRunEventSink()
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
del http_client
nonlocal model_requested
model_requested = True
@@ -2398,7 +2390,6 @@ def test_runner_rejects_multiple_output_layers_before_model_resolution(monkeypat
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -2431,7 +2422,7 @@ def test_runner_rejects_multiple_output_layers_before_model_resolution(monkeypat
)
sink = InMemoryRunEventSink()
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
del http_client
nonlocal model_requested
model_requested = True
@@ -2462,7 +2453,7 @@ def test_runner_rejects_reserved_output_name_with_wrong_layer_type_before_model_
) -> None:
model_requested = False
- def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient):
+ def fake_get_model(_self: DifyPluginLLMLayer, *, http_client: httpx.AsyncClient, agent_run_id: str):
del http_client
nonlocal model_requested
model_requested = True
@@ -2495,7 +2486,6 @@ def test_runner_rejects_reserved_output_name_with_wrong_layer_type_before_model_
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
RunLayerSpec(
@@ -2777,7 +2767,6 @@ def test_runner_treats_missing_runtime_dependency_as_validation_error() -> None:
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
]
@@ -2830,7 +2819,6 @@ def test_runner_treats_invalid_shell_snapshot_offsets_as_validation_error() -> N
plugin_id="langgenius/openai",
model_provider="openai",
model="demo-model",
- credentials={"api_key": "secret"},
),
),
]