Privacy policy

Jot is an open-source app that runs on your Mac. This policy covers the Jot app built from its public repository.

The short version

Your voice is transcribed on your Mac. No audio or text is sent to the makers of Jot or to any transcription service. Jot has no server, account, analytics, tracking, crash reporting or advertising.

Audio

• Dictation: while you dictate, the microphone audio is written to a recovery file in ~/Library/Application Support/Jot/Recordings/. When the dictation succeeds the file is deleted, unless you turn on Settings › "Keep audio of successful dictations". Failed, cancelled and interrupted dictations keep their audio so you can retry; delete them from History.
• Record Class and Record Call: audio is saved continuously to ~/Library/Application Support/Jot/Classes/ and kept until you delete the recording (it goes to the Trash). Record Call also captures what your Mac plays (the other side of a call) through a private Core Audio tap that exists only while recording. Jot shows no consent notice by design: recording people may require their consent where you live, so tell them.

Transcription: who receives your audio

Nobody. Speech is turned into text by models running on your Mac:

• Parakeet models (NVIDIA, Core ML conversions by FluidInference) through the FluidAudio library, on the Neural Engine.
• Apple's on-device speech (SpeechAnalyzer) for live words and class and call transcripts.

Text

• Transcripts, notes, folders, spelling rules and benchmarks are stored in ~/Library/Application Support/Jot/library.json; class and call transcripts next to their audio.
• Typing into other apps uses Accessibility to read the focused text field (to insert at the right place and to stop if you change it) and to write your words. When an app doesn't allow that, Jot pastes: the transcript is put on the clipboard and stays there.
• Cleanup runs on a model that Jot downloads and runs on your Mac; nothing is sent anywhere. If you instead choose Cloud in Settings and paste your own Groq, OpenAI, Anthropic or Gemini API key, each dictation's finished transcript text (never audio, notes or history) is sent in one HTTPS request to that provider, with your key in a request header. Redirects are refused. Cloud is off unless you turn it on.

Network requests

• Model downloads: when you download a model, Jot fetches its files from huggingface.co (the FluidInference repositories linked on each model card, and the Qwen3 4B Instruct file for cleanup, whose SHA-256 is checked before it's used). Nothing about you or your audio is sent.
• Apple speech assets: macOS may download its own on-device speech model the first time it's needed.
• Update checks: Jot checks newitools.app for updates about once a day. The request includes the app and macOS version. Only the number of checks per version is counted; nothing that identifies you is stored.
• Cloud cleanup: only if you add your own API key (above): the finished transcript text goes to the provider you picked.
• Nothing else.

Other local data

• ~/Library/Application Support/Jot/runtime.json: a small status file (ready or recording, window position, current page) rewritten every few seconds, for troubleshooting.
• While recording, Jot posts the distributed notification app.newitools.jot.recording (only "active: true/false") that other apps on your Mac can observe.
• Settings live in macOS user defaults: defaults delete app.newitools.Jot removes them.
• Deleting ~/Library/Application Support/Jot removes notes, history, recordings and models.

Changes

Changes to this policy are made in the repository's PRIVACY.md and visible in its history, and posted on this page with a new date.

Contact

Questions: hello@newitools.app, or open an issue on GitHub.