编排器 API

LangGraph-based orchestrator for switching between AI and human responses.

graph property

graph

Return the compiled LangGraph for external reuse and inspection.

__init__

__init__(
    *,
    model_name=None,
    model_provider=None,
    model_temperature=None,
    system_prompt=None,
    fallback_response=None,
    agent_record=None,
    model_routing_config=None,
    mcp_runtime_config=None,
    explicit_model_override_active=False,
)

Initialize the orchestrator.

参数：

名称	类型	描述	默认
model_name	str | None	Optional model name override.	None
model_provider	str | None	Optional provider override.	None
model_temperature	float | None	Optional temperature override.	None
system_prompt	str | None	Optional system prompt.	None
fallback_response	str | None	Optional fallback response override.	None

返回：

类型	描述
None	None.

引发：

类型	描述
Exception	Propagates unexpected initialization failures.

run

run(orchestrator_input)

Run a single orchestration step.

参数：

名称	类型	描述	默认
orchestrator_input	OrchestratorInput	Input payload.	必需

返回：

名称	类型	描述
OrchestratorResult	OrchestratorResult	Orchestration output.

引发：

类型	描述
Exception	Propagates unexpected failures from LangGraph execution.

Shared-infrastructure ReAct runtime for experimental chat agents.

__init__

__init__(
    *,
    model_name=None,
    model_provider=None,
    model_temperature=None,
    system_prompt=None,
    fallback_response=None,
    agent_record=None,
    model_routing_config=None,
    mcp_runtime_config=None,
    explicit_model_override_active=False,
)

Initialize the ReAct runtime.

run

run(orchestrator_input)

Execute one chat turn with the shared runtime kernel/pipeline.

Thin shared kernel that wraps event handling and root tracing.

__init__

__init__(
    *,
    tracing_service,
    apply_event,
    resolve_message_type,
    build_langfuse_metadata,
    record_observation_error,
)

Initialize the runtime kernel helper.

run_turn

run_turn(*, orchestrator_input, execute_turn)

Execute one orchestrator turn inside a shared root trace.

Run shared retrieval, memory, tool, generation, and observability stages.

__init__

__init__(*, agent_memory_service=None)

Initialize the shared capability pipeline.

run_turn

run_turn(
    *,
    request,
    retrieval_stage,
    skill_stage,
    scheduler_stage,
    mcp_stage,
    generation_stage,
)

Run the shared capability stages for one chat turn.

参数：

名称	类型	描述	默认
request	CapabilityPipelineRequest	Shared runtime payload for the turn.	必需
retrieval_stage	Optional[Callable[[CapabilityPipelineRequest, str], RetrievalStageResult]]	Optional deterministic retrieval stage callable.	必需
skill_stage	Optional[Callable[[CapabilityPipelineRequest, str], ToolStageResult]]	Optional skill execution stage callable.	必需
mcp_stage	Optional[Callable[[CapabilityPipelineRequest, str], ToolStageResult]]	Optional MCP execution stage callable.	必需
generation_stage	Callable[[CapabilityPipelineRequest, str], GenerationStageResult]	Final response generation stage callable.	必需

返回：

名称	类型	描述
CapabilityPipelineResult	CapabilityPipelineResult	Final turn response plus observability payload.

Input payload for the shared capability pipeline.

Final result returned by the shared capability pipeline.

Bases: BaseModel

Persisted resume state for one approval-gated MCP tool call.

Bases: BaseModel

Input payload for the orchestrator.

属性：

名称	类型	描述
session_id	str	Session identifier.
current_status	SessionStatus	Current session status.
incoming_role	str	Role of the inbound message.
incoming_content	str	Content of the inbound message.
conversation_history	list[ConversationHistoryMessage]	Prior session messages ordered oldest -> newest.
event_name	Optional[OrchestratorEvent]	Optional event to apply.
agent_id	Optional[str]	Agent ID for knowledge filtering in RAG retrieval.
persist_session_artifacts	bool	Whether session-bound memory, audit, and checkpoint records may be persisted for this turn.

Bases: BaseModel

Output payload after orchestration.

属性：

名称	类型	描述
next_status	SessionStatus	Status after processing the event.
response_role	Optional[str]	Role of the generated response.
response_content	Optional[str]	Content of the generated response.
observability_payload	Optional[dict[str, Any]]	Structured turn context metadata.
agent_id	Optional[str]	Agent ID used for this orchestration.

Bases: str, Enum

Represents the active responder for a session.

Bases: str, Enum

Represents events that can update the orchestration flow.