Subagents/workers die way too quick

[wet turtle]

I find my agents, subagents, workers dying off way too quick before they finish most long tasks. is there a way to change that?

[scenic moth]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[cloud surge]

Yep — there are two different “timeouts” that can make it look like agents/subagents/workers “die”.

1) Agent run gets aborted (common): agents.defaults.timeoutSeconds (default 600s)

By default, an agent run is aborted after ~10 minutes. Bump this if your worker runs are legitimately longer.

Config (JSON5):

{
agents: {
defaults: {
timeoutSeconds: 3600, // 1h (default: 600)
},
},
}

CLI way:

openclaw config set agents.defaults.timeoutSeconds 3600 --json
openclaw gateway restart

Docs: https://docs.openclaw.ai/concepts/agent-loop#timeouts

2) “Wait timed out” but the run continues (does NOT kill the worker)

If you’re dispatching work via sessions_send, the wait can time out and return status: "timeout" even though the agent is still running (you can check transcript/history later).

Docs: https://docs.openclaw.ai/concepts/session-tool#sessions_send

Sub-agents specifically: auto-archive (default 60 minutes)

Sub-agent runs are also subject to auto-archive (they can disappear from /subagents list after N minutes). If you run long subagent jobs, bump:

{

agents: {
defaults: {
subagents: {
archiveAfterMinutes: 240, // default: 60
},
},
},
}

Docs: https://docs.openclaw.ai/tools/subagents

---

If you tell me roughly how long they run before “dying” (30s vs ~10m vs ~60m) and what you see in /subagents list (outcome timeout / error / etc.), I can point you at the exact knob. Also grab openclaw --version (version mismatches matter a lot).