#Slow WebUI and interaction after 2026.4.25 update

1 messages · Page 1 of 1 (latest)

lean socket
#

Crazy slow and having gateway restarts after update to 2026.4.25 on Mac OS

fleet sirenBOT
#

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

hollow boneBOT
#

That sounds like a real post-update regression / startup loop, not normal slowness. 2026.4.25 is currently the npm latest, so first thing is to capture what’s looping.

Can you run/paste this?

openclaw status --all
openclaw gateway status --deep
openclaw doctor
openclaw logs --limit 300 --local-time

If logs are too noisy, look specifically for lines mentioning:

  • ENOTEMPTY
  • plugin-runtime-deps
  • openclaw-unknown
  • LaunchAgent
#
  • Bonjour / mDNS / ciao
  • “wrong version” / mixed version

Also worth checking for macOS split-brain after update:

which openclaw
openclaw --version
openclaw config get meta.lastTouchedVersion

If the binary/version/service don’t line up, the fix is usually reinstalling the LaunchAgent from the current install:

openclaw gateway install --force
openclaw gateway restart
lean socket
#

They match up

hollow boneBOT
#

Good — that rules out the mixed-version/split-brain path.

Next useful thing is the actual restart signature. Please paste:

openclaw status --all
openclaw doctor
openclaw logs --limit 300 --local-time

If doctor reports repairable plugin/runtime dependency issues, then run:

openclaw doctor --fix
openclaw gateway restart
#

For the log tail, the high-signal bits are usually errors around plugin-runtime-deps, ENOTEMPTY, LaunchAgent, Bonjour/ciao, or repeated “gateway starting” lines. Without that log chunk I’d just be guessing. 🦐

lean socket
#

There are no error showing but the it is painfully slow

hollow boneBOT
#

If there are no obvious errors, I’d treat this as a performance/stall diagnosis now.

Can you paste the output of these? They’ll show whether slowness is CLI→gateway, gateway internals, or Mac process thrash:

time openclaw status
time openclaw gateway status --deep
time openclaw health --verbose
openclaw status --all

And on macOS:

ps -axo pid,ppid,%cpu,%mem,etime,command | egrep 'openclaw|node|npm|tsserver|jiti' | sort -k3 -nr | head -30
#

Two quick sanity checks meanwhile:

  1. Try the Web UI in an incognito/private browser window, just to rule out stale frontend/cache weirdness.
  2. If doctor hasn’t been run with fixes since the update:
openclaw doctor --fix
openclaw gateway restart

The key question is: are openclaw status / health also slow, or only the Web UI? That splits the problem nicely.

young ingot
#

I concur that 4.25 is a lot slower with local LM studio Qwen3.6-27.
I also caught an error from LM studio logs stating model name 'url' field must be a base 64 encoded image.

hollow boneBOT
#

That LM Studio error is a really useful clue. It sounds less like “Web UI is slow” and more like LM Studio/Qwen is choking on OpenClaw’s richer OpenAI-compatible message shape — especially an image_url/structured content part.

For a text-only Qwen model, check the LM Studio model entry:

openclaw config get models.providers.lmstudio.models --json

Find the Qwen entry. If it has image input, make it text-only and add stricter compatibility flags. If it’s the first LM Studio model entry:

openclaw config set 'models.providers.lmstudio.models[0].input' '["text"]' --strict-json
openclaw config set 'models.providers.lmstudio.models[0].compat.requiresStringContent' true --strict-json
openclaw config set agents.defaults.experimental.localModelLean true --strict-json
openclaw gateway restart