Privacy

Local by default. Off-machine only when you say so.

Acousma is a desktop app, not a cloud service. There is no Acousma server, no account, and no telemetry pipeline for your meetings. The only bytes that can leave your computer are the ones you send to an AI provider you picked — and you can pick one that also runs locally.

Never leaves your machine

Your microphone audio — captured, transcribed, and discarded on your machine
Speaker separation — voice embeddings and clustering, all computed locally
Saved sessions — transcripts, notes, and voiceprints in %APPDATA%\Acousma on your disk
Your API keys — encrypted per-user with Windows DPAPI, never in plaintext, never logged
Settings and prompt templates — plain local files, no account, no cloud sync

Leaves only on your terms

When you use a hosted assistant

The context for that one request — transcript excerpts, plus any note bullets the assistant is expanding — is sent to the provider you chose (Anthropic, OpenAI, or Google) under your own API key. Nothing else, and never your raw audio.
When you point at a local server

Nothing leaves at all. Ollama, LM Studio, or any OpenAI-compatible server on your machine closes the loop — the whole app runs offline.

The short version

Transcription and speaker separation are fully offline. Your keys are encrypted on disk. Your own notes are never rewritten. The assistant is the one outbound call, it uses your provider and your key, it never carries your audio — and if you run a local model, there is no outbound call at all.

Questions about a specific data path? Ask by email.