1. Connect
Point the demo at a running server, sign in with an account, and let it mint a per-tab device key. Read the auth guide if any of that is new.
Not connected.
2. Upload & extract (server-side LLM)
Pick an image. The server runs OCR with an OpenAI vision model — no local OCR — using whatever settings you choose below. The result lands in the text box where you can edit it before pushing.
Extraction settings
3. Edit & push to /captures
Tweak what the AI extracted and store it. The image rides along, the
capture is keyed on clientId so retries don't duplicate.
AI-suggested form. Edit values, then push.
4. Latest captures
Pulled with GET /captures?limit=20. imageUrl
is a short-lived signed URL minted by the server.