how do I change this configuration agents.defaults.llm.idleTimeoutSeconds

[swift wind]

I do I change this configuration, im on the latest version of openclaw:
agents.defaults.llm.idleTimeoutSeconds

[peak ice]

Yep, that key still exists.

Use:

openclaw config set agents.defaults.llm.idleTimeoutSeconds 300 --strict-json

Then verify it:

openclaw config get agents.defaults.llm.idleTimeoutSeconds
openclaw config file

Notes:

• 300 = 5 minutes

• 0 = disable the LLM idle timeout
• active config is usually ~/.openclaw/openclaw.json

If you want, you can also validate after changing it:

openclaw config validate

Docs:

• https://docs.openclaw.ai/cli/config
• https://docs.openclaw.ai/concepts/agent-loop

If that command fails, paste:

openclaw --version

openclaw config get agents.defaults.llm.idleTimeoutSeconds

and I’ll help figure out what’s off.

[swift wind]

is there any disadvantage or anything I should know before setting it to 0 to disable the LLM idle timeout

[peak ice]

Yeah, there is one main tradeoff.

If you set agents.defaults.llm.idleTimeoutSeconds to 0, OpenClaw will stop aborting model calls that go silent mid-stream.

So the downside is:

• a stuck model/provider can sit there much longer
• runs can look “hung” instead of failing fast
• you’ll be relying on other limits, like agents.defaults.timeoutSeconds, provider timeouts, network disconnects, or manual cancel

So 0 is fine if you have a slow local model or a provider that legitimately pauses for a long time, but I usually would try 300 or 600 first before fully disabling it.

My recommendation:

• Cloud model: usually keep it enabled
• Slow local / heavy reasoning model: increase it first
• Only use 0 if you know the model regularly pauses and the timeout is the actual problem

If you do disable it, I’d make sure you still have a sensible outer timeout via agents.defaults.timeoutSeconds.