Local speech recognition
Raw system and microphone audio is handled by a local STT model in the desktop application. This reduces the data cloud components need.
- Parakeet for Russian and English
- GigaAM for Russian
- Whisper as a fallback model
Trust · Security
Security and data processing in Assist
Assist recognizes speech on your computer. To generate a response, conversation text and necessary context are sent to the Assist backend and an external AI provider.
Answer generation
To formulate an AI response, the app sends the current conversation text, question, and selected context to the backend and an external AI provider. Do not include information that should not be processed this way.
Permissions and window protection
The app needs microphone and system-audio permissions. Window capture limitation works only where the operating system supports it and does not guarantee invisibility in every screen-sharing application.
Your control
You start and stop sessions, manage the microphone, account, and top-up settings. The privacy policy and user agreements contain the full legal terms.
Frequently asked questions
Is raw audio sent to the AI provider?
Speech recognition runs locally. Text and necessary context are sent for answer generation instead of a raw audio stream for cloud transcription.
Does Assist guarantee the window stays out of screen sharing?
No. Assist uses operating-system capture-limitation mechanisms where supported, but behavior depends on the OS and sharing software.
Read the legal documents
The privacy policy explains the purposes and legal basis for data processing.
Open privacy policy