The meeting record is too important to rent
What is a self-hosted AI meeting notetaker? It is meeting AI that runs on infrastructure you control: Donna joins Google Meet from your own Linux VPS, records to your disk, transcribes via Whisper large-v3 with your API key, analyzes with your DeepSeek key, and stores transcripts and reports in your PostgreSQL — no vendor cloud holds any copy.
Every cloud notetaker asks the same quiet favor: let a startup keep a searchable archive of everything your company says in meetings. Boards, negotiations, one-on-ones, client calls — uploaded, processed, retained under someone else’s policy, exposed to someone else’s breaches. The favor is usually granted by default, one OAuth screen at a time. Donna is the refusal: same convenience, opposite custody.
What “self-hosted” means in Donna’s architecture
Capture on your hardware
A real Chromium browser driven by Playwright joins the call as a visible participant. Audio and screen video captured server-side with PulseAudio + ffmpeg, compressed after the call — the recording is born on your disk and never migrates.
AI on your keys
Transcription: Whisper large-v3 (via the Groq API, using your own key). Analysis: DeepSeek (your own key), two analysis passes. You hold the provider agreements; you can cap, audit, or replace them.
Storage you administer
PostgreSQL on your server; reports as Markdown; recordings on your disk. Retention is a cron job you write, not a plan tier you buy.
Access you configure
The dashboard sits behind your nginx authentication on your domain. Who reads meeting reports is a config file, not a SaaS sharing matrix.
Who needs this — not just prefers it
Teams whose meetings are their most sensitive artifact: boards handling MNPI, agencies under client NDAs, hiring panels holding candidate data, negotiators whose record is leverage, managers running honest one-on-ones, and anyone whose compliance posture treats third-party processors as attack surface.
The stack, plainly
Your own Linux VPS with nginx + pm2 — no vendor cloud in the data path. Xvfb provides the virtual display, PulseAudio the virtual audio device, ffmpeg the capture, Playwright the browser, PostgreSQL the memory, pm2 the persistence. No GPU required — the models are reached over API with your keys. Full detail: Can I run an AI notetaker on my own server?
Questions, answered straight
Is self-hosted the same as open source?
No — they answer different questions. Open source is about code visibility; self-hosted is about data custody. Donna is self-hosted commercial software: it runs entirely on your infrastructure with your keys. If open code is your hard requirement, see our honest comparison with Meetily.
Does any meeting data leave my server at all?
Two metered API calls do: audio chunks to Whisper large-v3 for transcription and transcript text to DeepSeek for analysis — both on your own keys, under agreements you hold directly. There is no third-party notetaker vendor in the path, and no copy is retained on our side because there is no “our side”.
Can I run Donna fully air-gapped?
Not today — transcription and analysis use external APIs on your keys. A local-model option would trade quality and cost; if your threat model requires full airgap, tell us: it shapes the roadmap.
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