Pinging CPU

[spark yarrow]

Gateway is extremely slow and am getting - gateway request timeout regularly.

I have done all the obvious things like prune sessions, restart gateway, restart machine, etc... We run a pretty deep agent setup with many running, but I don't think that is the issue necessarily. What could it be?

[placid pendant]

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

[ivory wraith]

I’m checking the docs + a local status baseline so I don’t guess.

I found a few likely culprits, especially around session-store churn and disk I/O. I’m pulling the specific docs so I can give you a sharper checklist instead of hand-waving.

Most likely, yes, but usually not because “many agents exist”. It’s often one of these:

1. Session-store / transcript churn
   Deep setups create lots of sessions.json writes, transcript files, thread sessions, and sometimes lock contention.
   Important: pruning session context is not the same as session-store cleanup.

2. Slow state storage
   If ~/.openclaw is on cloud sync, network storage, SD/eMMC, or just a slow disk, gateway RPC starts timing out.

3. Stuck bootstrapping
   A stray BOOTSTRAP.md in an agent workspace can keep an agent in first-run mode and make every turn heavier than it should be.

4. Local model/backend stalls
   Especially with local OpenAI-compatible backends, tiny direct calls can work while real agent turns stall on larger prompts/tool schemas.

Before guessing further, please paste these read-only checks, starting with version:

openclaw --version
openclaw status --all
openclaw doctor
openclaw sessions cleanup --dry-run
openclaw health --verbose

What I’d watch for in those outputs:

• bootstrapping agents > 0
• session-store count way above the default cap (500)
• orphan transcripts or session lock warnings
• warnings about slow/cloud/SD-backed state dir
• model/backend errors or repeated timeouts

Docs: