latent.agents.response¶

Structured agent response for conversation simulation.

Classes¶

AgentResponse¶

AgentResponse(text: str, tool_calls: list[ToolCallRecord] = list(), should_end: bool = False)

Response from an agent in a conversation simulation.

Attributes: text: The agent's text response. tool_calls: Tool calls made during response generation. should_end: Whether the agent signals the conversation should end.

StreamingAgent¶

StreamingAgent()

Protocol for agents compatible with respond_from_events.

Requires run() (async LLM call) and events (populated by run). Satisfied by ReActAgent and its subclasses.

ToolCallRecord¶

ToolCallRecord(tool_name: str, arguments: dict[str, Any] = dict())

A single tool call extracted from agent events.

Attributes: tool_name: Name of the tool that was called. arguments: Arguments passed to the tool.

Functions¶

respond_from_events¶

respond_from_events(agent: StreamingAgent, messages: list[Message], end_tool_name: str = END_CALL_TOOL_NAME) -> AgentResponse

Standard respond() implementation for agents in conversation simulation.

Calls agent.run(messages) (which clears and repopulates agent.events), then inspects events for tool calls and end-of-conversation signals.

Args: agent: A ReActAgent (or subclass) with run() and events. messages: Conversation history as Message objects. end_tool_name: Tool name that signals conversation end.

Returns: AgentResponse with text, typed tool calls, and end signal.

Methods¶

StreamingAgent.run¶

run(messages: list[Message]) -> str