What is the purpose of OPENCLAW_SKIP_STARTUP_MODEL_PREWARM and OPENCLAW_NO_RESPAWN

[marble kernel]

Because I have OPENCLAW_NO_RESPAWN = 1 commented in the code and OPENCLAW_SKIP_STARTUP_MODEL_PREWARM=1 and it worked because OC wasn't stable weeks ago but I still have them, I'd prefer knowing what are the +/-

[noble nest]

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

[fathom anvil]

Short version: I’d keep OPENCLAW_NO_RESPAWN=1 on small VPS/Pi/systemd-style installs, and I’d probably remove OPENCLAW_SKIP_STARTUP_MODEL_PREWARM=1 unless startup is still hanging/slow for you.

I checked current origin/main source:

OPENCLAW_NO_RESPAWN=1

• Avoids extra self-respawn / process handoff overhead.
• Keeps restart/PID behavior simpler on low-power hosts or custom supervisors.
• It’s explicitly recommended in the VPS startup tuning docs alongside NODE_COMPILE_CACHE.

• For update/full-restart paths, it can mean OpenClaw uses the in-process restart path instead of trying for a fresh PID/process handoff. Usually fine, but if you’re debugging “process state only clears after fresh restart” issues, temporarily unset it and do a real service restart.

OPENCLAW_SKIP_STARTUP_MODEL_PREWARM=1

• Skips the startup model prewarm, so gateway/channel startup can become less coupled to provider/model discovery.
• Useful if startup used to hang or slow down around model/provider probing.

• First real agent request may pay that cost instead.
• It can hide early warning signs that your default model/provider config is broken until the first interaction.
• Current source already wraps prewarm in a timeout and makes it metadata-ish, so this flag is less necessary than it may have been weeks ago.

[fathom anvil]

One nuance: version matters here. Please check:

openclaw --version

If you’re on a recent build and startup is stable, I’d test removing only OPENCLAW_SKIP_STARTUP_MODEL_PREWARM=1 first, restart, and watch:

openclaw logs --follow

Docs: https://docs.openclaw.ai/vps and FAQ: https://docs.openclaw.ai/help/faq

[marble kernel]

I'm on 2026.5.26

Ok so I'll keep the no respawn as my OC is in a docker container on pterodactyl. And I'll try without the skip model prewarm

[fathom anvil]

Yep, that’s the right split for a Docker/Pterodactyl setup.

Keep OPENCLAW_NO_RESPAWN=1: in a container, you generally want the panel/container supervisor to own process lifecycle instead of OpenClaw trying to hand off to another process.

Trying without OPENCLAW_SKIP_STARTUP_MODEL_PREWARM=1 makes sense on 2026.5.26. After restart, just watch for either a clean startup or a warning like startup model warmup failed / timeout in:

openclaw logs --follow

If startup stays clean, leave the prewarm skip removed. If it slows/hangs around model discovery, put that one back and it’s useful evidence for a GitHub issue.

[marble kernel]

Ok it seems to run and react

[fathom anvil]

Nice, that points to the prewarm skip no longer being needed for your setup.

If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution. That helps Answer Overflow index the answer for the next person with the same Docker/Pterodactyl question.