Documentation menu
DocsPrivacy and your data
Where your data lives, and what leaves your Mac
What Humla stores on your Mac, what it sends and to whom, and how to keep everything on the machine.
On this page
Humla has no backend of its own for your notes. Everything starts on your Mac, and something leaves it only when you have picked a service that needs it. This page lists what is stored where and what goes out.
What stays on your Mac
- Notes, transcripts and summaries are kept in one database on your Mac, at
~/Library/Application Support/no.humla.app/. - API keys and other secrets are kept in the macOS Keychain, never in that database. That includes your provider keys, your webhook address and headers, and your sign-in for Humla Cloud or GitHub Copilot.
- Speaker identification runs on your Mac, after a recording stops. So does removing the echo of a call from your microphone. No audio leaves your Mac for either.
- The MCP server that lets Claude Code or Codex read your notes runs on your Mac and reads the same database. It has no network access of its own.
Recorded audio
While you record, the audio is kept in a temporary folder and deleted once the transcript is finished and the speakers are identified. By default nothing is kept after that. A note still has its full transcript, but no player.
Turn on Settings → Recording → Keep recorded audio to keep recordings for playback and for identifying speakers again later. A recording then keeps:
- the audio you play back, mixed from both sides of the meeting
- your microphone and the system audio as separate files, which take roughly 2 MB per minute each
The setting applies to this Mac as a whole: with it off, a teammate's recordings aren't downloaded here either. Turning it off doesn't delete what is already stored. Delete stored audio… in the same place removes that, and keeps every transcript.
What leaves your Mac
Only what the services you choose need:
| What | Where it goes | When |
|---|---|---|
| Audio, in short clips | Your transcription provider: OpenAI, Deepgram, Groq or your own endpoint | While you record, unless you use on-device Whisper |
| Your notes and the transcript | Your summary provider: OpenAI, Anthropic, GitHub Copilot, your own endpoint, or a model server on your Mac | When you summarize a note, and when Humla gives a note its title |
| Your question and the parts of notes it finds | Your chat provider | When you use chat |
| Your notes, in short sections | OpenAI's embedding service, or the model server on your Mac | In the background as notes change, while chat is set to OpenAI or a local server, so chat can search by meaning. Not with Anthropic, GitHub Copilot or your own endpoint |
| A note, as JSON | Your webhook | Only when you choose Send to webhook… and confirm |
| Notes in a team workspace | Humla Cloud, hosted in the EU | When you use a team workspace |
Your API keys go to the provider they belong to and nowhere else.
A few things go out that aren't your notes:
- Update checks. Humla checks GitHub for a new version when it starts.
- Model downloads. On-device transcription and speaker models are downloaded once, from Hugging Face, the first time you need them.
- Anonymous counters. Humla counts first launch and whether setup was finished or skipped. They carry no identifier and nothing about your notes, and you can turn them off in Settings → General → Send anonymous usage counters.
Team workspaces
A team workspace syncs its notes through Humla Cloud, so your teammates have them too. Notes in your Personal space never leave your Mac.
- A note can be private, or shared with only the people you choose. Only you can change that for your own notes.
- Recording audio is uploaded to the workspace so teammates can play it back, only while Keep recorded audio is on. Turn off Upload recording audio in Settings → Account to keep the audio on your Mac; transcripts and summaries still sync.
- Chat in a team workspace runs on Humla Cloud, which searches the notes you are allowed to read.
Keep everything on your Mac
To make sure no audio or text ever leaves your Mac:
- Transcribe with on-device Whisper (Settings → Transcription).
- Summarize and chat with a model server on your Mac, such as Ollama or LM Studio (Settings → Summaries and Settings → Chat).
- Stay in your Personal space, and don't set up a webhook.
Speaker identification already runs on your Mac. What still goes out is the update check, the one-time model downloads, and the anonymous counters unless you turn them off.