Recording a call, transcribing it with Whisper and having a language model write the summary and action items is a sitting of work on your own machine, and a searchable archive with an ask-your-meetings box fits in the same session. The gap is the bot: MeetGeek's notetaker joins Zoom, Meet and Teams from your calendar, which means calendar OAuth, headless-browser or meeting-bot plumbing, per-person speaker labels and lobby rules that each platform changes. Record locally and skip the bot, and it's easy; keep the bot and it becomes a weekend plus maintenance.
Before you buildMeetGeek Basic gives 3 hours of transcription a month for nothing. If you have a few calls a week, you may not need to build anything.
Build me a local meeting recorder and note-writer to replace MeetGeek. Requirements: - Python 3.12 with FastAPI and Jinja2: `python -m notes` serves http://localhost:4710, and `notes record "Weekly sync"` starts a recording that Ctrl-C stops. - Capture: ffmpeg writes the microphone to the left channel and system audio to the right channel of one 16 kHz WAV in ./data/audio/. Also accept uploaded mp3, m4a, mp4 and wav files. - Transcription: `faster-whisper` with the model from .env (default `small`), word timestamps and automatic language detection. Print the model download size before the first run. - Speakers: the left channel is "You" and the right channel is "Others"; if HF_TOKEN is set, split "Others" with `pyannote.audio` into Speaker 1, Speaker 2 and so on. - Summary: any OpenAI-compatible endpoint from LLM_BASE_URL, LLM_MODEL and LLM_API_KEY in .env (default Ollama on localhost). Templates in ./templates/*.md (1:1, standup, sales call, interview) produce a summary, decisions, action items with owner, and open questions, each quoting the minute mark. - Storage: SQLite via `sqlite3` in ./data/notes.db with meetings, transcript segments and an FTS5 index; every finished meeting is also saved as ~/MeetingNotes/<date>-<title>.md. - Ask: a search box plus an "Ask your meetings" box that takes the top 8 FTS5 matches, sends them with their dates to the LLM, and answers with links to the meeting and timestamp. - Titles: read ICS_URL from .env (a secret iCal address) to name a recording after the event that is happening now. - Single user. No accounts, no telemetry, local-first; audio and text leave the machine only if LLM_BASE_URL points to a remote service. - Out of scope: a bot that joins calls, a mobile app, team spaces, CRM sync. - README: loopback setup per OS, rough transcription speed on CPU and GPU, and a reminder to tell attendees they're being recorded.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own MeetGeek.
Extend the meeting notes app in this folder. Requirements: - Store every action item as a row with owner, due date (if one was said), source meeting and status open or done; let me tick items off on the meeting page. - Add `notes digest` that emails the open items grouped by owner through SMTP settings in .env, and a cron example for Friday at 16:00. - Never invent an owner: if the transcript names nobody, leave it as "unassigned". - Update the README.
Add a pre-meeting reminder to the meeting notes app in this folder. Requirements: - A background process polls ICS_URL every 2 minutes and shows a desktop notification 1 minute before each event that has a video-call link, with a Start recording action. - Skip events whose title matches an ignore list in .env (for example "Lunch", "Focus"). - Name the recording after the event and attach the attendee list to the meeting page. - Update the README.
Add an Obsidian export to the meeting notes app in this folder. Requirements: - OBSIDIAN_VAULT in .env points at a vault; write each meeting to Meetings/<date> <title>.md with YAML frontmatter (date, attendees, tags) and wiki-links to people mentioned. - Re-running the export updates files in place and never touches notes I wrote by hand. - Add a `notes export --since 2026-01-01` command. - Update the README.
The pitch is that meetings take notes of themselves: the bot shows up, the summary lands in your inbox, and the team can search every call. People pay to stop remembering to press record and to have everyone's notes in one shared place.
Estimate · Whisper runs on your laptop and a local model writes the summaries for free; a hosted model for a handful of meetings a week costs a few dollars of tokens. The build is token usage on your agent plan.
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Basic | $0 | $0 | 3 hours of transcription a month, 2-hour meeting cap, transcripts kept 3 months, audio 1 month, no video. Bot and bot-free recording, uploads, 100+ languages, AI summary, AI chat and search, folders and tags, API and MCP access. |
| Pro | $15.99 | $9.99 | 20 hours of transcription a month, 2-hour meeting cap, transcripts kept 1 year, audio 6 months. Adds AI agents and workflows, Zapier, Make and n8n, unlimited integrations, meeting templates, personal analytics, downloads and exports. |
| Business | $27 | $17 | Unlimited transcription and transcript storage, 3-hour meeting cap, audio and HD video kept 12 months. Adds team spaces, comments, team analytics, private-by-default meetings, rules to skip recording, consent notices and custom branding. |
| Enterprise | custom | — | Business plus SSO and SCIM, org-wide locks, zero or custom data retention, on-premise storage options, 4-hour meeting cap, dedicated account manager. |
Basic is free: 3 hours of transcription a month, AI summaries, AI chat and search, API and MCP access.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: A notetaker that joins calls on its own from your calendar, Per-participant speaker names and the Zoom, Meet and Teams join handling.
About $15.99 a month on the Pro plan per seat, $191.88 a year (checked 2026-10-19).
A notetaker that joins calls on its own from your calendar; Per-participant speaker names and the Zoom, Meet and Teams join handling; Team spaces, comments and automatic sharing to the right people; Zapier, Make, n8n and CRM hand-offs after each call; Mobile and Chrome extension recording. If any of those matter to you, keep paying.
Yes: Meetily (open source), anarlog (open source), Vexa (open source).