CORS Chat
Simon Willison’s Weblog Simon Willison
Simon Willison built CORS Chat to test an OpenAI-style chat endpoint on local models. It works with LM Studio and OpenRouter, and even streams SVGs as they’re generated.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Simon Willison put together a new tool called CORS Chat today, using GPT-5.6-Sol xhigh, to make it easier to test Qwen 3.8 27B running in LM Studio on two machines: his M5 MacBook Pro and an NVIDIA DGX Spark.
The pitch is simple. It gives you a web interface for any chat endpoint that speaks the OpenAI Responses style of API. That makes it useful not just for one setup, but for poking at local and hosted model endpoints in the same way.
He says he tried it with LM Studio using the --cors option, and with OpenRouter too. Both worked.
There’s also a neat bit of polish here. The app keeps conversations in the browser, lets you export them as copy-pasted JSON, and watches for SVG images while they’re being generated so it can render them progressively before the stream is finished. That’s a very specific kind of nerd pleasure, and exactly the sort of thing that makes a tool feel built by someone who actually uses it.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of small tool: practical, a little opinionated, and aimed at making model testing less miserable. The AI world keeps selling grand platforms while the real work is still glue code, browser state, and annoying endpoint quirks. That’s where the useful stuff keeps showing up.
Read more about this at: Simon Willison’s Weblog
