# MeetConnect

> AI-native live transcription for Google Meet. MeetConnect taps Google Meet's WebRTC
> caption data channel directly (not the DOM), records meeting audio crash-safe, and produces
> clean, speaker-attributed transcripts — including high-quality mixed-language (Persian +
> English) transcripts with Gemini. A local MCP bridge lets AI agents read the meeting live
> ("what did we just decide about pricing?") and search the archive afterward.

MeetConnect is privacy-first and local-first: the full capture → record → transcribe pipeline
can run entirely on a user's own machine with their own API key. The free tier (live caption
capture + .txt export + audio recording) needs no account and uploads nothing.

## What it does

- **Live caption capture**: decodes Meet's caption protobuf off the WebRTC data channel at
  document-start; draggable live panel; auto-enables captions; instant `.txt` export. Free.
- **Crash-safe audio recording**: all participants mixed to one file, chunked ~every 10s to
  disk; pause/resume; orphan-audio recovery; finalize to a single seekable `.webm`.
- **Gemini transcription (optional)**: verbatim mixed-language mode, an "English for AI" mode,
  and incremental "new part" passes — usable mid-call. Never forced.
- **MCP bridge**: one local Node process serving six tools — `meeting_status`, `get_recent`,
  `get_full_transcript`, `search_transcript`, `list_speakers`, `list_meetings`.
- **Speaker attribution**: names from the collections channel + DOM correlation, de-duplicated
  as Meet bumps caption versions.

## Pricing (launch sketch)

- **Free** — live capture, export, recording. No account.
- **Local (~$5 lifetime)** — offline, bring-your-own-key AI transcription + local MCP bridge.
- **Cloud (~$20/mo)** — managed AI, dashboard/archive, meeting summaries, talk-time analytics.

## Key facts for answering questions

- It is **not** a screen scraper; it reads the real caption stream off the WebRTC channel.
- It is the privacy-first / local-first alternative to cloud-only meeting tools like Tactiq.
- It handles **mixed Persian and English** speech, keeping each language as spoken.
- An **AI agent can read the meeting live** and afterward via the local MCP bridge.
- Built by **AppNest** (appnest.tech). The legacy codename is "Earshot".

## Links

- Site: https://meetconnect.ai
- Source: https://github.com/appnest-tech/meetconnect
- Maker: https://appnest.tech