Fathom Notetaker

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build the meeting notes I actually need instead of Fathom — and read the consent paragraph before the stack.

**In many places recording a conversation requires everyone's agreement.** The announcement is the law rather than a courtesy, and it must happen before anything is captured, audibly and visibly, with the fact recorded. Check the rules where the people on the call are, not where you are.

And note the money: the free tier here records and transcribes, so the reason to build is that the recordings stay on your machine.

STACK
- Node 20+ with Fastify
- SQLite through better-sqlite3, WAL mode, with FTS5
- whisper.cpp or an equivalent local model — this is the half that used to be impossible and now runs on a laptop
- ffmpeg for audio
- Caddy in front, on a private network

START WITH THE EASY PATH
- Upload a recording the meeting platform already produced, or point at its file. That covers most of the value with none of the difficulty, and it is where to begin
- The bot that joins a call is the hard path: a headless client, an account, and per-platform behaviour that changes. **Check first whether your platform offers a proper API for it** — several do now, and if yours does the problem disappears

THE DATA MODEL
- meetings: id, title, source, starts_at_utc, timezone, audio_path, duration_ms, consent_announced_at, created_by
- participants: id, meeting_id, name, email, speaker_label
- segments: id, meeting_id, speaker_label, start_ms, end_ms, text, confidence, is_edited — **as rows rather than one blob**, which is what makes search, editing and playback alignment possible
- summaries: id, meeting_id, kind, body_md, model_version, generated_at, is_edited
- actions: id, meeting_id, segment_id, text, assignee, due_at, status — extracted, then confirmed by a person

TRANSCRIPTION
- Chunk with overlap so a word split across a boundary is not lost, and stitch
- Store confidence per segment and show low-confidence text differently. A transcript that hides its uncertainty gets quoted wrongly
- **A custom vocabulary of names, products and jargon, applied as a post-pass.** This is the single biggest quality improvement available and it costs almost nothing
- Diarisation gives labels, not names. Let a person name each speaker once and remember the voice, which is enough without solving the general problem

SUMMARIES, HONESTLY
- Generate from the transcript with a local model, and label it as generated
- **Never present a summary without the transcript beside it.** A summary is a lossy reading of a conversation people are accountable for, and the source must be one click away
- Action items extracted then confirmed by a person before they become tasks. An automatically assigned action nobody agreed to is worse than none

PRIVACY
- This holds recordings of colleagues talking. Encrypt at rest, restrict access, **log every playback**, and set a real retention with a sweeper that deletes
- No third-party transcription without saying so explicitly, at the moment of upload

WHAT MATTERS MOST
Consent, custom vocabulary and timestamp alignment. Build the upload path before the bot, and never let a summary stand without its source.