langchain-ai/deepagents · error · RuntimeError

agent invocation retry loop exited unexpectedly

Error message

agent invocation retry loop exited unexpectedly

What it means

_invoke_payload_with_retries loops retrying graph invocations and normally exits either by returning a result or by re-raising the last captured exception (last_exc). This RuntimeError is raised only if the loop ends with no result and no captured exception, which indicates an internal invariant violation (e.g. a break path that neither succeeded nor recorded an exception).

Source

Thrown at libs/talon/deepagents_talon/runtime.py:414

                return await invoke(payload, config=config)
            except asyncio.CancelledError:
                raise
            except Exception as exc:
                if not _is_retryable(exc) or attempt + 1 >= self.max_retries:
                    raise
                last_exc = exc
                backoff = min(2**attempt, 10)
                logger.warning(
                    "Retryable agent error in conversation %s; retrying in %ds: %s",
                    conversation_id,
                    backoff,
                    exc,
                )
                await asyncio.sleep(backoff)
        if last_exc is not None:
            raise last_exc
        msg = "agent invocation retry loop exited unexpectedly"
        raise RuntimeError(msg)

    def _graph_invoke(self) -> Callable[..., Awaitable[object]]:
        ainvoke = getattr(self._graph, "ainvoke", None)
        if not callable(ainvoke):
            msg = "Deep Agents graph does not expose async invocation"
            raise TypeError(msg)
        return cast("Callable[..., Awaitable[object]]", ainvoke)

    async def _invoke_until_unblocked(
        self,
        content: ModelContent,
        request: AgentRequest,
    ) -> object:
        state = await self._invoke_with_retries(content, request.conversation_id)
        for _ in range(DEFAULT_MAX_APPROVAL_ROUNDS):
            interrupts = _interrupts_from_state(state)
            if not interrupts:
                return state

View on GitHub (pinned to a1af029e6e)

Solutions

1. Report it to the library maintainers with a traceback and the runtime configuration
2. Check for subclasses/monkeypatches of _invoke_payload_with_retries or its retry hooks that break the loop invariant
3. Pin/upgrade to a version of the library where the retry loop is fixed

try:
    reply = await runtime.invoke(request)
except RuntimeError as exc:
    if "retry loop exited unexpectedly" in str(exc):
        log.exception("runtime invariant violated; retrying once with fresh runtime")
        runtime = rebuild_runtime()
        await runtime.start()
        reply = await runtime.invoke(request)
    else:
        raise

When it happens

Trigger: An internal edge case in the retry loop: the loop terminates without a successful payload and without last_exc being set. Not reachable through normal configuration.

Common situations: Effectively a bug guard; encountered after a library upgrade or custom subclassing that overrides retry hooks, or a corrupted retry configuration that skips the attempt body.

Related errors

• Local filesystem backend is unavailable.
• SDK {name} tool is unavailable.
• Received hook invocation interrupt without a HooksRuntime
• Unsupported hook event: {type(event).__name__}
• Unsupported hook-specific output: {type(specific).__name__}

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/e347cfb1c044aa4c. Report an issue: GitHub.