Noted

Quick start

1. Enter a fake Case ID — like PT-0001. Noted will prompt for patient initials + DOB on finalize; a hash-keyed case_id is generated automatically (e.g. PT-40F4656F4B). Never use real names.
2. Pick your Session type: Telehealth (video call) or In person.
3. Pick your ASR engine: Groq Whisper (recommended — higher accuracy, can capture the client's voice via tab sharing) or Browser native (Web Speech API; zero-latency mic only).
4. For telehealth, set Capture client audio to Mic + shared tab audio, then click your video service (Google Meet / Zoom / FaceTime / Webex / Teams / Doxy.me / SimplePractice / custom).
5. Press Start Session. Allow the mic. If tab audio is enabled, Chrome prompts you to pick the video-call tab — tick "Share tab audio" at the bottom-left of that dialog.
6. Speak. Captions appear live, newest at top, timestamped. Consecutive lines from the same speaker merge into one caption row. Chunks from the mic are tagged Therapist; chunks from the shared tab are tagged Client.
7. Use Mute for off-the-record moments (bathroom break, hold music). The mic stays live; captured audio is discarded until you press Unmute.
8. End Session → Finalize Note. The note appears in ~5 s (Groq) or 15–30 s (fallback). Pick from SOAP, DAP, BIRP, GIRP, PIE, or PAIP formats.

What's on the page

• Live captions (left column) — the session as it happens, scrollable, newest at top.
• Latest note panel (right column) — the most recent prior note for the current Case ID, auto-fetched as you type. Lets you reference continuity during the session.
• Cases page (nav above) — every case file in the vault, with session counts and an Obsidian-style index.
• Calendar card (on /cases) — Google Calendar events matched to case IDs when the event title/description contains PT-XXXX.

Where the data goes

• Groq Whisper — your audio chunks are sent to Groq's cloud for transcription. Groq does not sign BAAs, so this path is demo-only — don't use with real patient audio.
• Groq LLM (note generation) — your transcript + prior notes + psychology literature citations are sent to Groq for note drafting. Same demo-only caveat.
• Your Linode server — the generated note is written to /var/lib/noted/vault/<case_id>/ and kept for the cases page. Readable by Obsidian if you mount the folder as a vault.
• Google Calendar — read-only, uses your own OAuth client. Only reads event metadata to match case IDs.
• See Privacy & compliance for the full data-flow and HIPAA discussion.

Browser support

• Chrome / Edge desktop — full support including tab-audio capture for client voice.
• iOS Safari — mic + Browser-native engine work; tab-audio capture is not supported on iOS.
• Android Chrome — same as iOS Safari.
• Firefox / macOS Safari — Browser-native ASR is blocked; use Chrome or Edge.

Known limits

• Browser sandbox means tab-audio capture only hears the video-call tab if you pick it when starting the session. Desktop Chrome/Edge only.
• Web Speech API (browser native) caps sessions at ~60 s at a time; we auto-restart but there may be small gaps.
• On a 1 GB Linode, local LLM (Ollama) can't run an 8B model. Groq cloud is the default path.
• Production iOS / Android / Windows apps (on the roadmap) do everything on-device — no cloud, HIPAA-aligned by design.

Every field and button on this page has a small ? next to it — tap for plain-language help.