provenance-probe capture

Capture one AI chat request in your own browser and send it to your self-hosted provenance-probe instance for model analysis.

provenance-probe capture is a companion to a self-hosted provenance-probe instance — a tool that measures which AI model actually serves a chat app, and whether it is Chinese-origin or PRC-jurisdiction. Some AI chat apps present one brand while a different model answers behind the scenes. provenance-probe analyzes that. This extension is the easy way to feed it one real request: instead of exporting a HAR file by hand, you capture the request in your own browser, in your own logged-in session, and upload it to your own hosted instance. How it works: In the popup, set your instance URL + credentials once, and click Test connection. Open your AI chat app, log in, open DevTools → the "Provenance Capture" panel. Click Arm capture, send one short message, pick the request, and upload. Your instance runs its analysis and shows the result. What it does NOT do: No server-side browser and no new server surface — capture happens locally, in the tab you already have open. It never records your login (you log in yourself). It captures nothing until you explicitly Arm it, and only for the tab you're inspecting. It requests no broad permissions: no access to all sites, no reading your cookies or tabs. It asks for access to exactly the one instance host you type, only when you save. Your credentials are stored only in your browser (chrome.storage.local) and are sent only to the instance you configure, over HTTPS. Vendor API keys in a captured request are stripped before upload. Open source; see the repository for the full security write-up.