autogen_ext.runtimes.grpc#

class GrpcWorkerAgentRuntime(host_address: str, tracer_provider: TracerProvider | None = None, extra_grpc_config: Sequence[Tuple[str, Any]] | None = None, payload_serialization_format: str = 'application/json')[source]#

Bases: AgentRuntime

add_message_serializer(serializer: MessageSerializer[Any] | Sequence[MessageSerializer[Any]]) → None[source]#

Add a new message serialization serializer to the runtime

Note: This will deduplicate serializers based on the type_name and data_content_type properties

Parameters:: serializer (MessageSerializer[Any] | Sequence[MessageSerializer[Any]]) – The serializer/s to add

async add_subscription(subscription: Subscription) → None[source]#

Add a new subscription that the runtime should fulfill when processing published messages

Parameters:: subscription (Subscription) – The subscription to add

async agent_load_state(agent: AgentId, state: Mapping[str, Any]) → None[source]#

Load the state of a single agent.

Parameters:

agent (AgentId) – The agent id.
state (Mapping[str, Any]) – The saved state.

async agent_metadata(agent: AgentId) → AgentMetadata[source]#

Get the metadata for an agent.

Parameters:: agent (AgentId) – The agent id.
Returns:: AgentMetadata – The agent metadata.

async agent_save_state(agent: AgentId) → Mapping[str, Any][source]#

Save the state of a single agent.

The structure of the state is implementation defined and can be any JSON serializable object.

Parameters:: agent (AgentId) – The agent id.
Returns:: Mapping[str, Any] – The saved state.

async get(id_or_type: AgentId | AgentType | str, /, key: str = 'default', *, lazy: bool = True) → AgentId[source]#

async load_state(state: Mapping[str, Any]) → None[source]#

Load the state of the entire runtime, including all hosted agents. The state should be the same as the one returned by save_state().

Parameters:: state (Mapping[str, Any]) – The saved state.

async publish_message(message: Any, topic_id: TopicId, *, sender: AgentId | None = None, cancellation_token: CancellationToken | None = None, message_id: str | None = None) → None[source]#

Publish a message to all agents in the given namespace, or if no namespace is provided, the namespace of the sender.

No responses are expected from publishing.

Parameters:

message (Any) – The message to publish.
topic (TopicId) – The topic to publish the message to.
sender (AgentId | None, optional) – The agent which sent the message. Defaults to None.
cancellation_token (CancellationToken | None, optional) – Token used to cancel an in progress. Defaults to None.
message_id (str | None, optional) – The message id. If None, a new message id will be generated. Defaults to None. This message id must be unique. and is recommended to be a UUID.

Raises:

UndeliverableException – If the message cannot be delivered.

async register_factory(type: str | AgentType, agent_factory: Callable[[], T | Awaitable[T]], *, expected_class: type[T] | None = None) → AgentType[source]#

Register an agent factory with the runtime associated with a specific type. The type must be unique. This API does not add any subscriptions.

Note

This is a low level API and usually the agent class’s register method should be used instead, as this also handles subscriptions automatically.

Example:

from dataclasses import dataclass

from autogen_core import AgentRuntime, MessageContext, RoutedAgent, event
from autogen_core.models import UserMessage


@dataclass
class MyMessage:
    content: str


class MyAgent(RoutedAgent):
    def __init__(self) -> None:
        super().__init__("My core agent")

    @event
    async def handler(self, message: UserMessage, context: MessageContext) -> None:
        print("Event received: ", message.content)


async def my_agent_factory():
    return MyAgent()


async def main() -> None:
    runtime: AgentRuntime = ...  # type: ignore
    await runtime.register_factory("my_agent", lambda: MyAgent())


import asyncio

asyncio.run(main())

Parameters:

type (str) – The type of agent this factory creates. It is not the same as agent class name. The type parameter is used to differentiate between different factory functions rather than agent classes.
agent_factory (Callable[[], T]) – The factory that creates the agent, where T is a concrete Agent type. Inside the factory, use autogen_core.AgentInstantiationContext to access variables like the current runtime and agent ID.
expected_class (type[T] | None, optional) – The expected class of the agent, used for runtime validation of the factory. Defaults to None.

async remove_subscription(id: str) → None[source]#

Remove a subscription from the runtime

Parameters:: id (str) – id of the subscription to remove
Raises:: LookupError – If the subscription does not exist

async save_state() → Mapping[str, Any][source]#

Save the state of the entire runtime, including all hosted agents. The only way to restore the state is to pass it to load_state().

The structure of the state is implementation defined and can be any JSON serializable object.

Returns:: Mapping[str, Any] – The saved state.

async send_message(message: Any, recipient: AgentId, *, sender: AgentId | None = None, cancellation_token: CancellationToken | None = None) → Any[source]#

Send a message to an agent and get a response.

Parameters:

message (Any) – The message to send.
recipient (AgentId) – The agent to send the message to.
sender (AgentId | None, optional) – Agent which sent the message. Should only be None if this was sent from no agent, such as directly to the runtime externally. Defaults to None.
cancellation_token (CancellationToken | None, optional) – Token used to cancel an in progress . Defaults to None.

Raises:

CantHandleException – If the recipient cannot handle the message.
UndeliverableException – If the message cannot be delivered.
Other – Any other exception raised by the recipient.

Returns:

Any – The response from the agent.

start() → None[source]#: Start the runtime in a background task.

async stop() → None[source]#: Stop the runtime immediately.

async stop_when_signal(signals: Sequence[Signals] = (Signals.SIGTERM, Signals.SIGINT)) → None[source]#: Stop the runtime when a signal is received.

async try_get_underlying_agent_instance(id: ~autogen_core._agent_id.AgentId, type: ~typing.Type[~autogen_ext.runtimes.grpc._worker_runtime.T] = <class 'autogen_core._agent.Agent'>) → T[source]#

Try to get the underlying agent instance by name and namespace. This is generally discouraged (hence the long name), but can be useful in some cases.

If the underlying agent is not accessible, this will raise an exception.

Parameters:

id (AgentId) – The agent id.
type (Type[T], optional) – The expected type of the agent. Defaults to Agent.

Returns:

T – The concrete agent instance.

Raises:

LookupError – If the agent is not found.
NotAccessibleError – If the agent is not accessible, for example if it is located remotely.
TypeError – If the agent is not of the expected type.

class GrpcWorkerAgentRuntimeHost(address: str, extra_grpc_config: Sequence[Tuple[str, Any]] | None = None)[source]#

Bases: object

start() → None[source]#: Start the server in a background task.

async stop(grace: int = 5) → None[source]#: Stop the server.

async stop_when_signal(grace: int = 5, signals: Sequence[Signals] = (Signals.SIGTERM, Signals.SIGINT)) → None[source]#: Stop the server when a signal is received.

class GrpcWorkerAgentRuntimeHostServicer[source]#

Bases: AgentRpcServicer

A gRPC servicer that hosts message delivery service for agents.

async GetState(request: AgentId, context: ServicerContext[AgentId, GetStateResponse]) → GetStateResponse[source]#: Missing associated documentation comment in .proto file.

async OpenChannel(request_iterator: AsyncIterator[Message], context: ServicerContext[Message, Message]) → Iterator[Message] | AsyncIterator[Message][source]#: Missing associated documentation comment in .proto file.

async SaveState(request: AgentState, context: ServicerContext[AgentId, SaveStateResponse]) → SaveStateResponse[source]#: Missing associated documentation comment in .proto file.