Can I run an AI notetaker on my own server?
Yes. A standard Linux VPS can run the full pipeline: an automated browser joins the meeting, a virtual audio device and ffmpeg record it, Whisper transcribes the audio, and an LLM writes the report. Donna packages exactly this stack — Playwright, PulseAudio, ffmpeg, Whisper large-v3, DeepSeek, PostgreSQL — as one self-hosted product.
What does the server actually need?
Less than you would guess: a Linux VPS with a virtual display (Xvfb), a virtual audio device (PulseAudio), ffmpeg, and Node.js. Donna’s setup script installs the stack, and pm2 keeps the processes alive. No GPU is required, because the heavy models are reached via API — Whisper large-v3 through Groq with your key, DeepSeek for analysis with yours. The server’s jobs are attending, recording, compressing, and storing, which ordinary VPS hardware does comfortably.
Why not just use a free cloud notetaker?
Three structural reasons. Custody: cloud tools keep a copy of every meeting on their servers; your own server keeps the only copy. Cost shape: per-seat subscriptions scale with headcount, while one server attends for the whole team and API usage is metered in pennies per meeting-hour. Control: your retention policy is a command on your own disk, not a plan feature. The trade is that you administer a server — which for teams already running nginx and pm2 is a familiar Tuesday.
What does Donna add over DIY scripts?
The join choreography (lobby, admission, meeting-end detection), crash-safe capture, post-call compression, chunked transcription of long recordings, a two-pass analysis that goes far beyond summarization, a dashboard with live status and full report viewing, an upload path for past recordings, and Postgres persistence — the difference between a weekend proof-of-concept and a colleague you can rely on.
Related questions
Do I need a GPU to self-host meeting transcription?
Not with Donna’s architecture — Whisper large-v3 runs via the Groq API on your key, so the VPS only records and orchestrates. If you require fully local inference, GPU-backed local Whisper is possible but is a different cost and quality trade.
How much server do I need?
An ordinary VPS of the class teams already rent for web apps: enough disk for compressed recordings and enough CPU for ffmpeg compression after calls. There is no per-seat scaling — one server serves the whole team.
Is maintenance heavy?
It is a standard Node service under pm2 behind nginx: the same shape as any small production app. Updates are a git pull and a build; the database is a single PostgreSQL instance with migrations included.
Put Donna in your next meeting
Donna deploys onto your own VPS in an afternoon: nginx, pm2, PostgreSQL, your API keys. Early access is open — tell us about your team and we’ll get her a seat at your table.
Request early access