You're right to be concerned — that error means there's a circular wait between the tool and the speech handle. I’m responding to your question about whether to await an

AgentTask

from inside a

function_tool

and how to avoid the RuntimeError. Short answer • Don’t await an

AgentTask

inside a

function_tool

. That can create circular waits when the task or agent triggers speech. • If you need to wait for the agent’s spoken audio to finish inside a tool, use

RunContext.wait_for_playout()

(i.e.,

await context.wait_for_playout()

), not

SpeechHandle.wait_for_playout()

. Why this happens •

function_tool

runs as part of the tool’s execution context. If the agent/task you await produces speech, awaiting the task can indirectly cause a

SpeechHandle

that the tool also tries to wait on — producing the circular wait shown in your error. What to do instead • If you only need to wait for speech to finish inside the tool: ◦ Use

await context.wait_for_playout()

inside the

function_tool

. • If you’re trying to run an

AgentTask

and wait for its result: ◦ Run the

AgentTask

from the agent-level context (e.g., agent

on_enter

or other agent code) and await it there, not from the tool. • If the task must be initiated from the tool, refactor so the tool signals the agent to create/run the task, then the agent awaits it (tool → signal → agent runs/awaits task). If you’d like, share the minimal snippet showing how you create/await the task and any `say`/speech calls; I can point out the exact lines causing the circular wait and suggest a concrete refactor. Sources: Tasks and task groups | LiveKit docs | Agent speech and audio | LiveKit docs