#Openclaw has become unusable since 2026.4.14 updates.

1 messages · Page 1 of 1 (latest)

marble bear
#

Using unbuntu OS / GPT 5.4 codex - The context window warning is showing max 200k tokens not the default 274k, compaction hardly ever works, heart beats interrupt any task thats running, messages i send hang and i only see responses if i refresh the chat - sometimes no response at all. these errors just persist through updates and just seem to get worse. Starting to think I may just be better cutting my losses and switching to claude code

weak roostBOT
#

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

olive hearthBOT
alpine eagleBOT
#

Let me check the docs and current references so I don’t guess.

#

That sounds genuinely rough, and no, that’s not “normal enough to live with.”

A couple of concrete points first:

  • If you’re actually on openai-codex/gpt-5.4, 200k is not the stock runtime cap. The docs currently say the default runtime cap is 272k, so 200k usually means something is overriding contextTokens or you’re not on the model/provider path you think you are.
  • Heartbeats should normally defer when the main lane is busy, not stomp active work.
  • “Replies only appear after refresh” sounds more like a UI/WebSocket/client sync problem than a pure model problem.

Can you paste these, in this order:

openclaw --version
openclaw status --all
openclaw gateway status
openclaw models status
openclaw logs --follow
#

Then reproduce one “hang” and paste the relevant log chunk.

Also tell me:

  1. the exact model ref you’re on (openai-codex/gpt-5.4 vs codex/gpt-5.4 vs openai/gpt-5.4)
  2. where this is happening, dashboard web chat, TUI, or somewhere else
  3. whether heartbeat is enabled/customized

Two quick mitigations you can try right now:

openclaw system heartbeat disable

If that immediately improves stability, heartbeat config is part of the problem.

And if you want heartbeat without blowing up context, the docs recommend:

#

I’d avoid that deskdiscord.com link, it’s not an OpenClaw support path.

If this needs maintainer attention, the right escalation is a GitHub issue:
https://github.com/openclaw/openclaw/issues

Danny, if you want to get this taken seriously as a bug, please paste:

openclaw --version
openclaw status --all
openclaw gateway status
openclaw models status
openclaw logs --follow

Then reproduce one hang and include that log chunk.

#

Also, if you’re on openai-codex/gpt-5.4, 200k is not the normal default cap. That usually means an override or a different provider/runtime path than expected.

marble bear
#

openclaw --version
OpenClaw 2026.4.21 (f788c88)

openclaw status --all
OpenClaw status --all

Overview
Version 2026.4.21
OS linux 6.17.0-20-generic (x64)
Node 22.22.2
Config ~/.openclaw/openclaw.json
Dashboard http://127.0.0.1:18789/
Tailscale off
Channel stable (default)
Update pnpm · up to date · npm latest 2026.4.21
Gateway local · ws://127.0.0.1:18789 (local loopback) · reachable 47ms
· auth token · <host> (<lan-ip>) app 2026.4.21 linux 6.17.0-20-generic
Gateway self unknown
Gateway service systemd installed · enabled · running (pid <pid>, state
active)
Node service systemd not installed
Agents 1 total · 0 bootstrapping · 1 active · 5 sessions
Secrets none

Channels
none

Agents
main bootstrap file: ABSENT sessions: 5 active: 3m ago
store: ~/.openclaw/agents/main/sessions/sessions.json

Diagnosis
Gateway target: ws://127.0.0.1:18789
Config: ~/.openclaw/openclaw.json
Bind: loopback

âś“ Config: ~/.openclaw/openclaw.json
âś“ Secret diagnostics (0)
âś“ Restart sentinel: none
! Port 18789 is already in use
- pid <pid> <host>: openclaw-gateway (127.0.0.1:18789)
- Gateway already running locally. Stop it (openclaw gateway stop) or use a
different port.
! Tailscale: off · unknown
✓ Skills: 13 eligible · 0 missing · ~/.openclaw/workspace
âś“ Plugin compatibility (none)
âś“ Channel issues (none)

#

Gateway restart attempts tail:
[2026-04-22T07:45:50Z] openclaw restart attempt source=update target=openclaw-
gateway.service
[2026-04-22T07:45:50Z] openclaw restart done source=update

openclaw gateway status
Service: systemd (enabled)
File logs: /tmp/openclaw/openclaw-2026-04-22.log
Command: /usr/bin/node ~/.npm-global/lib/node_modules/openclaw/dist/index.js
gateway --port 18789
Service file: ~/.config/systemd/user/openclaw-gateway.service
Service env: OPENCLAW_GATEWAY_PORT=18789

Config (cli): ~/.openclaw/openclaw.json
Config (service): ~/.openclaw/openclaw.json

Gateway: bind=loopback (127.0.0.1), port=18789 (service args)
Probe target: ws://127.0.0.1:18789
Dashboard: http://127.0.0.1:18789/
Probe note: Loopback-only gateway; only local clients can connect.

Runtime: running (pid <pid>, state active, sub running, last exit 0, reason 0)
Connectivity probe: ok
Capability: admin-capable

Listening: 127.0.0.1:18789

openclaw models status
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : codex/gpt-5.4
Fallbacks (0) : -
Image model : -
Image fallbacks (0): -
Aliases (0) : -
Configured models (2): codex/gpt-5.4, openai-codex/gpt-5.4

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (1): openai-codex (1)

  • openai-codex effective=profiles:~/.openclaw/agents/main/agent/auth-
    profiles.json | profiles=1 (oauth=1, token=0, api_key=0) | openai-
    codex:default=OAuth
#

Missing auth

  • codex openclaw configure or set an API key env var.

OAuth/token status

  • openai-codex usage: 5h 96% left ⏱3h 27m · Week 99% left ⏱6d 16h
    • openai-codex:default ok expires in 34h

openclaw logs --follow
Captured short recent excerpt instead of an endless stream:

09:25:07+00:00 info gateway ready (6 plugins: acpx, browser, device-pair,
memory-core, phone-control, talk-voice; 7.2s)
09:25:16+00:00 info webchat connected client=openclaw-control-ui webchat
v2026.4.21
09:25:16+00:00 info chat.history UNAVAILABLE during gateway startup
09:25:16+00:00 info models.list UNAVAILABLE during gateway startup
09:25:28+00:00 info chat.history âś“
09:32:13+00:00 info webchat disconnected code=1001
09:32:13+00:00 info webchat connected client=openclaw-control-ui webchat
v2026.4.21
09:32:14+00:00 info commands.list âś“
09:32:14+00:00 info models.list âś“
09:35:52+00:00 info channels.status âś“

Older relevant log lines:
08:21:02+00:00 401 Unauthorized for codex/gpt-5.4
08:46:14+00:00 invalid chat.history params: must have required property
'sessionKey'; unexpected property 'sessionId'

alpine eagleBOT
#

Thanks, that helps.

A few useful reads from that output:

  • You’re on current stable.
  • The gateway is up and reachable.
  • That Port 18789 is already in use line is just the running openclaw-gateway, not the smoking gun.
  • Channels: none means the “I only see replies after refresh” issue is probably local dashboard/web chat state, not channel delivery.

The next things I’d check are:

openclaw models status
openclaw config get agents.defaults.contextTokens
openclaw config get agents.defaults.heartbeat
openclaw config get models.providers.openai-codex
# 
openclaw config get models.providers.codex

Then run:

openclaw logs --follow

Keep that open, send one message that hangs, then paste the relevant log lines.

Two fast isolation tests:

  1. Disable heartbeat temporarily:
# 
openclaw system heartbeat disable
  1. Start a fresh chat/session with:
/new

If a fresh session behaves better, that points more at session/compaction state than the whole install.

Also, if the dashboard only updates after refresh, try a hard refresh or another browser once, just to separate UI/WebSocket weirdness from backend/model issues.

Big clue to watch for:

  • if agents.defaults.contextTokens is 200000, that explains the 200k warning immediately.
  • if it’s unset, then I want to see models status, because you may not actually be on the provider/runtime path you think you are.
#

I’ve got the key clue now, let me verify the Codex-side defaults so I don’t tell you the wrong thing.

#

Nice, this is the missing clue.

You’re not on openai-codex/gpt-5.4, you’re on codex/gpt-5.4, which means you’re using the Codex app-server harness.

That matters a lot because:

  • codex/gpt-5.4 = Codex harness path
  • openai-codex/gpt-5.4 = OpenClaw PI path with Codex OAuth

Also, the bundled Codex provider defaults to 272k, so 200k still does not look like the normal default. That makes me suspect either:

  • agents.defaults.contextTokens is set to 200000, or
  • you’ve got stale session/runtime reporting, or
  • the Codex harness/app-server path is the thing going sideways.

Next commands I want:

# 
openclaw config get agents.defaults.contextTokens
openclaw config get agents.defaults.embeddedHarness
openclaw config get plugins.entries.codex

And in the affected chat/session:

/codex status

If that works, also try:

/codex compact
#

Reason: on the codex/* path, compaction is delegated to the Codex app-server, not the normal PI compaction path.

Best A/B test from here:

  1. Start a fresh session with /new
  2. Switch that session to:
/model openai-codex/gpt-5.4
  1. Retry the same workflow

If the hangs / refresh-only replies / compaction weirdness disappear on openai-codex/gpt-5.4, then the problem is very likely in the Codex harness/app-server path, not OpenClaw globally.

I’d still also do the quick heartbeat isolation once: