Free and open source software enthusiast, Maintainer and active contributor of few projects.
Taiwan, R.O.C
Joined December 2014
- Tweets4.1K
- Following3K
- Followers728
- Likes5.3K
Teamed up with Codex + GPT-6 Astra today to dig into the LiteLLM v1.107.0 PrivCollect compromise 🔍
Built a toolkit with IoCs, custom YARA rules, an offline scanner, and response guidance.
First try. I'm no malware expert. Feedback and PRs welcome!
github.com/PeterDaveHello/li…
If you need proper invoices for AI credits, be careful with @OpenRouter.
Their billing may produce wrong invoice details. I filed a support ticket, but 1+ month later there's still no corrected invoice or ETA. Refunding to repurchase also means the processing fee isn't refunded.
Hey @OpenRouterAI, the docs say openrouter/fusion-flash is a separately listed model with its own /api/v1/models entry, but API calls currently return a 400 error: fusion-flash is not a valid model ID.
Are the docs ahead of the rollout, or is this a bug?
Doc: openrouter.ai/docs/guides/ro…
Find the best AI models and hardware for you.
local.ai/peterdavehello/invi…
Small Codex CLI gripe: review mode saved review_rollout_user as a message ID even though the API rejects it. Compaction kept copying it through the history. On resume, HTTP 400 made a two-month, 500 MB session unusable, so I had to patch the JSONL by hand to recover it. 😮💨
Review mode created the bad ID, compaction propagated it into nested history, and any follow-up, resume, or remote compact that replayed it could hit the same 400 because message IDs must start with msg.
Related reports: github.com/openai/codex/issu… and github.com/openai/codex/issu…
Zhipu AI (Z.ai) learned how to reset the usage limit 🙂
ZCode now has 1 million users. As a thank-you to our community, we’ve reset usage limits for all GLM Coding Plan users.
We’re also rolling out an update that helps turn long-horizon capabilities into completed work:
- More intelligence in real engineering workflows
- 98% cache hit rate, providing around 1.8x more usage
Download: zcode.z.ai/en
Join the community: discord.gg/EpH5XkTyhu
ChatGPT and Codex are both erroring out right now.
503 Service Unavailable
auth error code: biscuit_baker_service_me_circuit_open
The breaker is open on the auth service, so requests aren't even reaching the backend. Retrying just burns time.
Anyone else getting this in the last few minutes?
Back to fully operational. ChatGPT, Codex, and the API are all responding again on this end 🎉
Thanks to the OpenAI team for the quick turnaround, and to everyone who chimed in while it was down.
Mitigation is in, and OpenAI is now monitoring the recovery.
Things should gradually start coming back, though some requests may still fail for now 😅
