#switch from claude to openai for openclaw
1 messages · Page 1 of 1 (latest)
Let me verify the actual docs paths so I don't hand-wave config keys.
The FAQ path moved a bit in this checkout, I’m locating the right one now.
Yep, easiest path:
If you want direct OpenAI API usage (openai/*):
openclaw onboard --auth-choice openai-api-key
openclaw models set openai/gpt-5.4
openclaw models fallbacks clear # optional, only if you want to stop Claude fallbacks too
openclaw models status
If you want ChatGPT/Codex sign-in instead (openai-codex/*):
openclaw onboard --auth-choice openai-codex
openclaw models set openai-codex/gpt-5.4
openclaw models fallbacks clear
openclaw models status
A couple gotchas:
- You don’t need to delete your Anthropic key unless you want to.
- If it still uses Claude, check whether you still have an Anthropic fallback configured.
- The default model key is
agents.defaults.model.primaryif you’re editing config manually.
Docs:
- https://docs.openclaw.ai/providers/openai
- https://docs.openclaw.ai/concepts/models
- https://docs.openclaw.ai/help/faq
If you want, paste:
openclaw --version
openclaw models status
and I’ll tell you exactly what to change.
does 5.5 work?
how do i get 5.5 to work? currently it is only 5.4 installed and saying 5.5 is unavailable.
Let me check whether 5.5 is actually in the current OpenClaw catalog or if this is just a version gap.
If you’re using the normal OpenAI API key route (openai/*), that’s probably expected.
Current docs say:
openai/gpt-5.5only works once OpenAI exposes GPT-5.5 on the public API- Right now, GPT-5.5 access is via Codex auth in OpenClaw:
openai-codex/gpt-5.5
So to use 5.5 now, switch to the Codex route:
openclaw models auth login --provider openai-codex
openclaw models set openai-codex/gpt-5.5
openclaw models status
Or with onboarding:
openclaw onboard --auth-choice openai-codex
If you want to stay on OPENAI_API_KEY, then you’ll need to stick with an API-exposed model like 5.4 until OpenAI enables 5.5 there.
Please paste these two so I can tell whether it’s just the route, or also a version/setup issue:
openclaw --version
openclaw models status
Docs:
I can try to clean this up now. I’ll first fetch the thread messages so I can remove the sensitive ones instead of guessing.
Safest cleanup is deleting the whole support thread if Discord lets me. I’m checking that first, because it’s one action instead of trying to hunt every pasted log line.
I can’t delete the whole thread, so I’m trying message-by-message cleanup instead.
I’m removing the newer pasted diagnostics and mirrored bot replies first, then I’ll check what’s left.
I’ve removed the big diagnostic dump range. I’m checking what’s still left so I can tell you exactly what remains.
Done. I removed the pasted logs/status dumps and the bot replies that mirrored them.
What’s still left is just the early non-sensitive part of the thread, plus the auto welcome messages. I couldn’t delete the whole thread because the bot doesn’t have that permission.
If you want, I can also remove the remaining non-sensitive messages I’m allowed to delete, but the thread itself would still need a mod/admin delete.