AgentAsToolOptions

@open-agent-loops/core

type AgentAsToolOptions = 
  | AgentAsToolIdentity & {
  agent: Agent;
}
  | AgentAsToolIdentity & Omit<RunAgentOptions, "prompt" | "sessionId" | "memory" | "signal"> & {
  agent?: undefined;
  memory?: Memory;
  run?: RunFn;
};

Defined in: tools/agent-as-tool.ts:90

Options for agentAsTool — two mutually exclusive forms.

Config form reuses the run contract: everything runAgent needs except the fields the tool controls — prompt (the model's task), sessionId and memory (the isolation defaults), and signal (forwarded from the tool-call context). So model is required, and system / tools / maxSteps / stopWhen / hooks / toolExecution / onEvent configure the child run. Mirrors the Omit<RunAgentOptions, …> "RunBase" shape used by runGoal and the dispatcher.

Agent form takes one self-contained Agent value instead. The agent closes over its own model and memory, so per-call context isolation becomes the agent's concern: an agent whose memory persists will remember across calls (each call still gets a distinct default sessionId, so isolation holds unless you also pin AgentAsToolIdentity.sessionId).