Local app (source)
The same engine as the desktop app, run from source. Use it on a PC with an NVIDIA GPU for fast Whisper, or on a home server behind a private network so a few colleagues can open the page.
Install
unzip meeting-ear-local-app.zip
cd meeting-ear
pip install -e core
meetingear devices
meetingear config mic="<your mic name>" engine=claude-code
meetingear serve --open
Python 3.11 or newer. On a GPU machine also pip install nvidia-cublas-cu12 nvidia-cudnn-cu12. Claude Code installed and logged in for the claude-code engine, or meetingear config engine=api --api-key sk-ant-....
The page is at http://127.0.0.1:8771/. The compact strip is at /compact.
Briefs
briefs/ next to the code: one folder per client. Copy briefs/_template to start one. connectors.json in the folder declares live sources; meetingear sync <client> writes their facts into the brief before a call.
Test without a call
meetingear replay <client> docs/sample-transcript.txt --speed 4 --open
Sharing it on a private network
Run meetingear serve --host 0.0.0.0 on the server and put it behind your VPN or Tailscale. The page has no login of its own, so never expose it to the public internet.
Commands
| Command | What |
|---|---|
meetingear serve | start the server and page |
meetingear devices | list microphones and outputs |
meetingear sync <client> | pull facts from the client's connectors |
meetingear connectors <client> | show connectors and their tools |
meetingear replay <client> <file> | replay a transcript |
meetingear config key=value | settings: engine, api_model, action_model, mic, loop, whisper_model, budget_usd, auto_recap |