Recommended LLM timeout setting
✅Solved
I ran into an issue when Openclaw would take 15-30 minutes to respond. My sysadmin AI gave me this diagnosis and proposed solution: lower the timeout from 600 seconds to 120 seconds. Is that a good idea? Can OpenClaw tell the difference between an LLM query that is working and taking a long time, versus a query that is simply not responding / hung?
---
The permanent-memories agent experienced 15-30 minute delays last night due to a combination of an API outage and a high default timeout.
The logs from last night show: [agent/embedded] embedded run timeout: ... timeoutMs=600000 [diagnostic] lane task error: lane=session:agent:permanent-memories:telegram:group:-5169748248 durationMs=600078 error="FailoverError: LLM request timed out."
Root Cause:
At around 1:30 AM local time, requests to the primary google:default profile started hanging (likely an API outage or extreme rate-limiting on Google's end).
OpenClaw's default LLM request timeout (agents.defaults.timeoutSeconds) is 600 seconds (10 minutes).
Every single request waited 10 entire minutes before giving up.
After giving up, it failed over to the next model in your 6-item fallbacks list.
As the sequence progressed, you experienced 15-30+ minutes of waiting before the agent could successfully reply (or definitively fail).
It resolved on its own this morning because the primary Google API recovered and the group's massive context window was automatically flushed by the idleMinutes: 30 setting.
Proposed Changes
We should lower the global timeoutSeconds so OpenClaw fails over much faster when an API hangs.
openclaw.json
Will modify the agents.defaults block to explicitly set timeoutSeconds to 120 (2 minutes). 2 minutes is a healthy balance—long enough to process massive Second Brain contexts (550k+ tokens), but short enough to quickly fail over if the API genuinely hangs.
---
The permanent-memories agent experienced 15-30 minute delays last night due to a combination of an API outage and a high default timeout.
The logs from last night show: [agent/embedded] embedded run timeout: ... timeoutMs=600000 [diagnostic] lane task error: lane=session:agent:permanent-memories:telegram:group:-5169748248 durationMs=600078 error="FailoverError: LLM request timed out."
Root Cause:
At around 1:30 AM local time, requests to the primary google:default profile started hanging (likely an API outage or extreme rate-limiting on Google's end).
OpenClaw's default LLM request timeout (agents.defaults.timeoutSeconds) is 600 seconds (10 minutes).
Every single request waited 10 entire minutes before giving up.
After giving up, it failed over to the next model in your 6-item fallbacks list.
As the sequence progressed, you experienced 15-30+ minutes of waiting before the agent could successfully reply (or definitively fail).
It resolved on its own this morning because the primary Google API recovered and the group's massive context window was automatically flushed by the idleMinutes: 30 setting.
Proposed Changes
We should lower the global timeoutSeconds so OpenClaw fails over much faster when an API hangs.
openclaw.json
Will modify the agents.defaults block to explicitly set timeoutSeconds to 120 (2 minutes). 2 minutes is a healthy balance—long enough to process massive Second Brain contexts (550k+ tokens), but short enough to quickly fail over if the API genuinely hangs.
Solution
- Do reduce the timeout if you care about responsiveness during outages (120–300s depending on how “big” your typical turns are).
- Reduce/reorder fallbacks so you don’t have 6 long waits in a row (especially if multiple fallbacks share the same upstream failure mode).
- If you want, paste:
-
- the relevant
…and I’ll suggest a concrete value + a safer per-agent split.
- Reduce/reorder fallbacks so you don’t have 6 long waits in a row (especially if multiple fallbacks share the same upstream failure mode).
- If you want, paste:
-
openclaw --version- the relevant
agents.defaults.timeoutSeconds…and I’ll suggest a concrete value + a safer per-agent split.
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
129,194Members
Resources
Was this page helpful?
