---
title: Agent Stuck in a Retry Loop
summary: An agent repeatedly retries the same failed tool call instead of stopping or escalating: usually an expired credential or an autonomy setting mismatch.
date: 2026-08-20
---

# Agent Stuck in a Retry Loop

An agent repeatedly retries the same failed tool call instead of stopping or escalating: usually an expired credential or an autonomy setting mismatch.

If an agent appears to be repeating the same action over and over without making progress, it's almost always retrying a tool call that's failing the same way each time, not a bug in the agent's reasoning itself.

## Most common cause: an expired or revoked credential

If the integration a tool call depends on has an expired token that failed to refresh, or was revoked from **Settings → Integrations**, every retry fails identically. Reconnect the integration from its settings card. See the [Integrations Guide](/en/docs/guides/integrations-guide) for how token refresh normally works automatically.

## Rate limits on the external service

A third-party API rate limit (not a HeartBeatAgents limit) can look identical to a stuck loop from the outside. The agent keeps retrying a call the external service keeps rejecting. Check that service's own dashboard for rate-limit or quota errors around the same timestamps.

## Autonomy level mismatch

An agent in **Autonomous** mode has no natural point to stop and ask for help when a tool call fails. It just keeps trying. If a task genuinely needs a human decision partway through, moving that agent to **Supervised** (see [Agent Management](/en/docs/guides/agent-management)) gives it a place to stop and surface the failure instead of retrying blindly.

## If none of the above applies

Check the agent's conversation log for the actual tool error being returned: the retry loop is a symptom, and the underlying tool error usually names the real cause directly.

## Related

- [Agent Management](/en/docs/guides/agent-management): autonomy levels and when to use Supervised mode
- [Security](/en/docs/guides/security): how credentials are scoped and rotated