Privacy policy
Last updated October 20, 2026
The short version. Your meetings, notes, transcripts and recordings stay on your computer unless you send them somewhere yourself. Trailmix sends a few anonymous usage stats once a day, so it can support the computers people actually use. They never include anything from your meetings, your name or your files, and you can turn them off at any time.
What stays on your computer
Everything Trailmix records or writes is kept on your computer, in Trailmix's own folder: recordings, transcripts, notes, action items, names, workspaces, tags, your answers to its quick questions, and your settings (including any API keys you add). Transcription runs on your computer unless you choose Google Gemini for it, and so do the notes when you use a local AI model through Ollama. Recordings are deleted after 30 days unless you choose to keep one, or after the time you pick in Settings → Privacy (as soon as they're transcribed, after 7 days, or never).
Recording other people
Trailmix is a tool, and you decide what to record with it. Many places require everyone on a call to agree before it's recorded or transcribed, and data-protection laws may apply when you record for work. Following the law where you and the people you record are is up to you. To help, Trailmix shows a short summary of the rules for your country or US state during setup and in Settings → Privacy. It's general information, not legal advice, and it changes no setting unless you choose to apply one.
Anonymous usage stats
Trailmix is free. To decide which computers and set-ups to support, it sends a few anonymous stats. They're on by default, and Trailmix asks before sending anything: the first time you open it, there's a box to untick. You can turn them off at any time in Settings → Privacy, where See what's sent shows the data exactly as it goes out.
What's sent:
- Once a day: the Trailmix version; your operating system and version; your kind of computer (chip, number of cores, memory, disk size and free space, rounded); which main features are on; which AI writes your notes (for example "Ollama, qwen2.5:7b" or "Claude"); and rough counts of meetings and workspaces, in ranges like "6–20".
- After a meeting is processed: its length (rounded to 5 minutes), whether it was recorded or imported, how long transcription and notes took, and which AI wrote the notes.
- When processing fails: the stage and the kind of error, never the error message, because messages can contain names or file paths.
- When Trailmix updates, or you go back to an earlier version: the two version numbers.
What's never sent: anything from your meetings (titles, notes, transcripts, names, workspaces, tags, recordings), your name, your API keys, file names or paths, or the addresses of your servers.
How it's kept anonymous: each copy of Trailmix uses a random ID, not linked to you, your account or your computer's identity; turning the stats off forgets it, and turning them back on starts a new one. The stats are stored with PostHog, an analytics service, as anonymous events: no personal profile is built, no location is looked up, and IP addresses are discarded. They're kept no longer than needed to support Trailmix, and at most two years.
What Trailmix connects to
- GitHub, to check for updates, download them, and fetch an updated ranking of AI models and the recording-laws summary from this website. GitHub sees your IP address, as with any website.
- Hugging Face, once, to download the speech models; and Ollama's library if you download a local AI model through Trailmix.
- A cloud AI you add (such as Claude, OpenAI, Gemini or DeepSeek). If you choose one to write your notes, your transcript is sent to it for that, under that company's own privacy policy. Trailmix never adds one for you.
- Your own transcription server, if you set one up, which receives your recordings.
- Google Gemini for transcription, only if you choose it (in setup or Settings → Transcription). With live captions on, your microphone and the call's audio stream to Google while you record, silence excluded. After you stop, the recording is uploaded (your two tracks mixed into one unless you turn that off) to be transcribed with the speakers told apart, and Trailmix deletes the upload as soon as it's done. Redoing old transcripts, and the cheaper batch jobs for transcripts and notes, send audio or transcripts to Google the same way, only when you ask. It's all under Google's own privacy policy and your API key's terms. Lockdown mode switches every part of this off.
- Problem reports are only sent if you post them: Trailmix removes personal information on your computer, shows you the report, and you post it yourself as a public issue on GitHub.
This website
This website counts visits with GoatCounter: no cookies, no personal data, and no tracking across sites. It's hosted on GitHub Pages, which, like any web host, sees your IP address, and the download button asks GitHub which release to offer.
Your choices
You can turn the usage stats off at any time in Settings → Privacy. Because they're anonymous, past stats can't be traced back to you or your copy of Trailmix once the ID is forgotten. Where the law gives you rights over your data (for example in the EU or UK), the stats are sent on the basis of a legitimate interest in keeping Trailmix working on people's computers, and you can object at any time by turning them off.
Changes and questions
If this policy changes, the new version is posted here with a new date, and anything that changes what Trailmix sends is described in that version's release notes. For questions, open an issue on GitHub.