Consented study log. One tap starts a session: this phone records audio, a live transcript (when the browser allows it), and location metadata until you stop. A visible REC indicator stays on. No hidden mic.

Conversations are split after 60 seconds of silence. Live captions are a preview only. When a conversation closes, Field transcribes the saved audio with OpenAI (gpt-4o-mini-transcribe, else whisper-1) if you have set a key, then understands the text with gpt-4o JSON (proofreading stays on gpt-4o-mini). Without a key, on-device regex analysis is kept as a labeled fallback.

Audio stays in IndexedDB on this device. Transcripts are sent to OpenAI only when an API key is set in Settings (field:openaiKey).

No account. No Field server. Export JSON from Settings (audio blobs stay on the phone).

• iPhone: Safari, Share, Add to Home Screen. Keep Field open while recording (iOS suspends background capture in a PWA).
• Android: browser menu, Add to Home screen.

Live captions may use Apple or Google speech services. Prefer Chrome for on-screen transcript.

Vite base is ./ so a project site works. CI builds on push to main.