openclaw - 💡(How to fix) Fix [Bug]: openai-codex/gpt-5.4 authenticates successfully but reply fails with “DNS lookup for the provider endpoint failed” [1 participants]

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Hello OpenClaw team,

We are running OpenClaw 2026.4.14 on Ubuntu 24.04.4 LTS, installed globally with npm, in local gateway mode on a self-hosted machine. The dashboard is opened from a Windows PC through an SSH tunnel to 127.0.0.1:18789.

Current setup:

• Provider: openai-codex
• Model: openai-codex/gpt-5.4
• OAuth login completes successfully
• openclaw config get agents.defaults.model.primary -> openai-codex/gpt-5.4
• openclaw gateway status --require-rpc -> RPC probe: ok

What works:

• Gateway starts correctly
• Dashboard connects
• Logs show webchat connected
• Internal calls work: commands.list, chat.history, models.list, node.list, device.pair.list

Problem: When we send a message from the dashboard, the reply never comes back. Logs show: provider=openai-codex error=LLM request failed: DNS lookup for the provider endpoint failed

Checks already done:

• Re-authenticated Codex OAuth from scratch
• Removed old auth-profiles and moved ~/.codex/auth.json away
• Forced transport to "sse" in ~/.openclaw/openclaw.json
• Validated openclaw.json successfully
• Restarted gateway multiple times
• Host DNS works
• Node DNS + fetch tests also work
• chatgpt.com/backend-api is reachable

Because both host-level and Node-level DNS/network tests succeed, this looks like a provider-specific OpenClaw issue rather than a real DNS failure on the machine.

Could you please advise what to check next, or whether this is a known issue/regression with openai-codex in 2026.4.14?

Thanks.

Steps to reproduce

1. Install and run OpenClaw 2026.4.14 on Ubuntu 24.04.4 LTS using npm global.
2. Configure the default model as openai-codex/gpt-5.4.
3. Authenticate successfully with Codex OAuth using: openclaw models auth login --provider openai-codex --set-default
4. Restart the gateway and confirm: openclaw config get agents.defaults.model.primary -> openai-codex/gpt-5.4 openclaw gateway status --require-rpc -> RPC probe: ok
5. Open the dashboard through an SSH tunnel to 127.0.0.1:18789 from a Windows PC.
6. Send a simple prompt such as: "Responde solo: OK".
7. Observe that no reply is returned and the gateway logs show: provider=openai-codex error=LLM request failed: DNS lookup for the provider endpoint failed

Expected behavior

The dashboard should return a normal model response to the prompt after successful OAuth authentication and with the gateway running normally.

Actual behavior

The dashboard connects and accepts the message, but no reply is returned. Gateway logs show: provider=openai-codex error=LLM request failed: DNS lookup for the provider endpoint failed

OpenClaw version

2026.4.14

Operating system

Ubuntu 24.04.4 LTS (host), dashboard opened from Windows 11 through SSH tunnel

Install method

npm global

Model

openai-codex/gpt-5.4

Provider / routing chain

OpenClaw -> openai-codex/oauth -> openai-codex/gpt-5.4

Additional provider/model setup details

• OAuth login completes successfully.
• Default model is set to openai-codex/gpt-5.4.
• Gateway status returns RPC probe: ok.
• Dashboard is accessed through SSH tunnel to 127.0.0.1:18789.
• We forced transport to "sse" in ~/.openclaw/openclaw.json.
• We removed ~/.openclaw/agents/main/agent/auth-profiles.json and re-authenticated.
• We also moved ~/.codex/auth.json away and repeated OAuth.
• The issue persists after restart and fresh authentication.

Logs, screenshots, and evidence

Observed log lines:
- webchat connected
- commands.list, chat.history, models.list, node.list, device.pair.list succeed
- provider=openai-codex
- error=LLM request failed: DNS lookup for the provider endpoint failed

Network and DNS checks already done:
- getent hosts auth.openai.com -> OK
- getent hosts api.openai.com -> OK
- getent hosts chatgpt.com -> OK
- curl -I https://auth.openai.com -> OK
- curl -I https://api.openai.com -> OK
- curl -I https://chatgpt.com/backend-api/ -> reachable, returns HTTP 403
- Node dns.lookup() for chatgpt.com, auth.openai.com and api.openai.com -> OK
- Node fetch("https://chatgpt.com/backend-api/") -> status 403

This suggests host-level and Node-level DNS/network are working, while the failure appears inside the OpenClaw openai-codex provider path.

Impact and severity

Affected: Self-hosted OpenClaw deployment using openai-codex on this machine Severity: High, because chat replies cannot be generated Frequency: 100% in all tested attempts Consequence: Dashboard messages are accepted but no model response is returned

Additional information

NOT_ENOUGH_INFO