openclaw - 💡(How to fix) Fix [Bug]: 2026.4.29 Web UI chat bubbles then disappears; direct OpenAI API works [4 comments, 4 participants]

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Hi OpenClaw Support,

I’m having an issue where OpenClaw loads normally, but the Web UI chat does not return any assistant response.

System / environment

• OpenClaw version: 2026.4.29
• macOS: 26.2 arm64
• Node: 22.22.2
• Gateway: local loopback at 127.0.0.1:18789
• Provider: OpenAI
• Model tested: gpt-4.1-mini
• Heartbeat: disabled
• Skills/hooks/channels/search: skipped/minimal during fresh setup

What happens

• The dashboard opens normally.
• A new session starts.
• I type a simple prompt like: “Reply only: OK”
• The assistant bubble appears/loading starts.
• Then the bubble disappears and no response is returned.

What I already tried

• Restarted gateway multiple times.
• Disabled heartbeat.
• Switched from GPT-5.5 / GPT-5 mini to GPT-4.1 mini.
• Cleared stuck sessions.
• Performed a fresh uninstall/reinstall.
• Skipped skills, hooks, channels, and search provider.
• Tested the OpenAI API directly outside OpenClaw with curl using gpt-4.1-mini, and it returned “OK” successfully.

Relevant log patterns

• openai/gpt-4.1-mini produced no reply before the idle watchdog
• embedded run failover decision ... reason=timeout
• FailoverError: LLM request timed out
• stuck session ... processing_with_queued_work
• Earlier TUI issue: getWindowSize ENOTTY
• Occasional Web UI disconnects / token mismatch from stale dashboard tabs

Since the direct OpenAI API call works, this looks like an OpenClaw embedded agent/runtime/Web UI issue rather than an OpenAI billing/API-key issue.

Can you help me identify the cleanest fix or the exact diagnostic bundle/logs you need?

Thanks, Mostafa

Steps to reproduce

1)Install or update to OpenClaw 2026.4.29. 2)Configure OpenClaw with OpenAI as the provider. 3)Use gpt-4.1-mini as the model. 4)Skip skills, hooks, channels, and search provider for a minimal setup. 5)Open the Web UI dashboard at http://127.0.0.1:18789/. 6)Start a new chat/session. 7)Send a simple prompt: Reply only: OK. 8)The assistant loading bubble appears, then disappears with no assistant response.

Expected behavior

OpenClaw should return a normal assistant response in the Web UI. For the test prompt Reply only: OK, it should respond with OK.

Actual behavior

The Web UI opens and the chat session appears to start, but when I send a message, the assistant loading bubble appears briefly and then disappears. No assistant response is returned.

Direct OpenAI API testing outside OpenClaw works. I tested gpt-4.1-mini with curl and received OK, so the OpenAI key/model/API access appear to be working.

Logs show repeated embedded agent timeouts and stuck processing behavior, including:

openai/gpt-4.1-mini produced no reply before the idle watchdog embedded run failover decision ... reason=timeout FailoverError: LLM request timed out stuck session ... processing_with_queued_work

Earlier TUI also froze/crashed with:

getWindowSize ENOTTY TUI exited with code 1

OpenClaw version

2026.4.29

Operating system

macOS 26.2 arm64 Node 22.22.2 Mac mini

Install method

No response

Model

openai/gpt-4.1-mini Also previously tried: openai/gpt-5-mini openai/gpt-5.5

Provider / routing chain

Provider: OpenAI Routing: OpenClaw embedded agent → OpenAI provider profile → openai/gpt-4.1-mini Gateway: local loopback ws://127.0.0.1:18789 Dashboard: http://127.0.0.1:18789/ Heartbeat: disabled Skills/hooks/channels/search: skipped/minimal during fresh setup

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

Direct OpenAI API test outside OpenClaw succeeds. Using curl against the OpenAI Responses API with gpt-4.1-mini returned OK immediately. The failure appears to be in OpenClaw’s embedded agent/Web UI/runtime path rather than OpenAI API access or billing.