Loading
A third-party LLM provider sees your prompts.
Impossible by construction - Typilot does not call any third-party LLM. All inference is Ollama on your machine.
A model fine-tunes on your data.
No data ever leaves the device, so no model can be trained on it. You can verify by inspecting Ollama logs.
Telemetry leaks your prompts to analytics.
No prompt-level telemetry. The only outbound call is a license check; you can review traffic with Little Snitch or a packet inspector.
Meeting audio uploaded for processing.
Whisper runs locally with Metal acceleration on Apple Silicon. Audio files stay in your local SQLite store; you choose to keep or delete them.
Cloud sync of notes, history, meetings.
There is no cloud sync. All artifacts are in a SQLite file in your user data directory.
Independent assessment confirms no personal data leaves the device. Article 5(1)(c) minimisation principle satisfied.
BSI-aligned review of local-only inference, telemetry boundaries, and update integrity. No server-side processing.
Consumer right-to-know report: zero collected personal information. No sale or sharing of user data.
No. All AI inference runs on your machine through Ollama, Whisper, and Supertonic. The only network call by default is a license check.
On macOS, Little Snitch or LuLu will show every outbound connection. On Linux, nethogs or tcpdump. Typilot will only show the license endpoint.
No. Audio is captured locally, transcribed locally with Whisper, and stored in a local SQLite database. You can delete the raw audio after the summary is generated.
Your prompts, notes, meetings, and history live in the app data directory. Uninstall removes the app; you can manually delete the data directory to remove everything.
License verification logic is source-available so you can confirm exactly what data - and only what data - is sent over the network.