Openclaw slow v2026.4.25

[brittle dock]

My delay is before the model starts:

• Inbound recorded: 22:24:14.912
• Agent session.started: 22:25:45.479
• Delay before agent/model: about 90.6s
• Model completed: 22:25:48.015
• Model time: about 2.4s

[vagrant anchor]

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

[grand dome]

Yep — that timing means not the model/provider. The delay is happening before the embedded agent run reaches session.started.

Most likely causes:

1. Queue/debounce config — especially if something is near 90000ms.
2. Main lane saturated — inbound turns wait behind other active main turns/heartbeats.
3. Gateway restart/runtime-deps drain — also has 90s-ish paths in some cases.

Can you run:

openclaw --version
openclaw config get messages.queue
openclaw config get messages.inbound
openclaw config get agents.defaults.maxConcurrent
openclaw config get cron.maxConcurrentRuns

If you see debounceMs: 90000 or debounceMsByChannel.discord: 90000, that’s the smoking gun.

For the queue case, during/after a repro:

openclaw logs --limit 800 --local-time --plain | grep -E 'lane wait exceeded|queueDepth|message queued|draining|runtime deps|session.started'
openclaw gateway stability --json

If logs show lane wait exceeded: lane=main waitedMs≈90000, raise/check agents.defaults.maxConcurrent — default should be 4, so if it’s 1, try:

openclaw config set agents.defaults.maxConcurrent 4
openclaw gateway restart

Docs: https://docs.openclaw.ai/concepts/queue and https://docs.openclaw.ai/gateway/diagnostics

[brittle dock]

i think this issue is with the new update

[grand dome]

Yeah, I’m leaning that way too. There are already fresh 2026.4.25 regression issues open, including one around deferred gateway restarts leaving inbound sessions stuck, and another around gateway startup blocking on pricing fetch timeouts. Your timing — 90s before session.started, then 2.4s model time — points at Gateway/queue/reload plumbing, not the model. (github.com)

Can you grab this from the affected machine?

openclaw --version
openclaw logs --limit 1500 --local-time --plain | grep -E 'reload|deferred|lane wait exceeded|stuck session|pricing|runtime deps|session.started'
openclaw gateway stability --json

If you see restart still deferred, stuck session, or lane wait exceeded, that’s strong evidence. The queue path is relevant because inbound replies are intentionally queued by session + global main lane before the agent starts. (docs.openclaw.ai)

If it’s reproducible on 2026.4.25, please file/add to a GitHub issue with those log lines and exact version. Happy to help interpret the output here.

[brittle dock]

which the fastest and most stable openclaw previous version

[grand dome]

Assuming you’re on 2026.4.25, I’d roll back to 2026.4.23 right now.

Why: 2026.4.24 also had some suspicious regression reports, so 2026.4.23 is the safer “last known calmer” target than just one step back.

Try:

openclaw --version

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --install-method npm --version 2026.4.23

openclaw doctor
openclaw gateway restart
openclaw --version

If you’re running from source instead: