My tool call is being interrupted, but the agent is not able to pickoff from where it was interrupted

Thanks for reaching out, Aumesh. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

but after the interruption shouldn't the agent be able to decide whether to call the function tool again or not?

You're right to expect the agent to be able to decide — but it won't do that automatically. When a tool call is interrupted, the call is cancelled and removed from history; the agent itself must detect that and choose whether to call the tool again. What you can do • Detect the interruption inside the tool using

context.speech_handle.interrupted

. That lets you clean up and signal the agent that the operation ended prematurely. • Persist partial progress (database, cache, session state) so on next turn the agent can inspect state and decide whether to retry, continue from a checkpoint, or abandon. • Use explicit retry logic in the agent: after an interruption, have the agent ask the user whether to retry or automatically re-invoke the tool based on session/state rules. • Prevent interruptions for critical sections with

context.disallow_interruptions()

when you need the operation to run to completion. Minimal code example for detection and cleanup

@function_tool()
async def my_long_running_task(self, context: RunContext):
    wait_for_result = asyncio.ensure_future(self._long_operation())
    await context.speech_handle.wait_if_not_interrupted([wait_for_result])

    if context.speech_handle.interrupted:
        wait_for_result.cancel()
        # persist any partial state here if useful
        return None
    else:
        return await wait_for_result

Suggested flow to let the agent decide 1. Tool runs and writes checkpoints/partial results to storage. 2. If interrupted, tool returns/cleans up (and optionally notes interruption in session state). 3. On the next agent turn, check session/state: ◦ If state shows an incomplete operation, either prompt the user (“Resume previous task?”) or automatically re-invoke the tool with resume parameters. ◦ If no resume desired, clear the incomplete state. If you tell me what type of tool and what state you can store (DB, Redis, in-memory), I can suggest a concrete retry/resume pattern or sample code tailored to your case.