Local voice on Windows

A private voice assistant for Windows

Boris listens, transcribes, and speaks on your own PC. Only the text of what you ask, and the context needed to answer it, goes to the AI model you pick. This page lists every place your data goes.

What runs on your PC

Wake word
A local ONNX classifier listens for the wake word. Audio is scored in memory and never uploaded.
Speech-to-text
NVIDIA Parakeet transcribes your request locally. Raw microphone audio stays on your PC.
Voice output
Supertonic speech synthesis turns replies into audio locally.
Memory and sessions
Facts, preferences, transcripts, and logs are stored under your Boris home folder in a local SQLite database and plain files.

After the speech models are installed, wake detection, transcription, and spoken replies keep working without an internet connection. Boris has no analytics or usage telemetry, and ONNX Runtime telemetry is switched off.

What leaves your PC

Your AI model
The transcribed text of your request, plus relevant conversation, memory, and tool results, goes to the model you choose through OpenRouter. Their privacy policy applies to that request.
Web search
When Boris searches the web, the query goes to Exa if you added an Exa key, or to DuckDuckGo otherwise. Pages it reads are fetched from their sites.
Downloads and updates
Speech models download once from Hugging Face. Update checks contact GitHub Releases.

You choose the model and the provider. OpenRouter lets you pin a provider, and model usage is billed to your own account.

Stay in control

API keys are stored in plain text in auth.json inside that folder, so protect your Windows account. Approval prompts and path rules reduce risk, but Boris is not a full sandbox. Read the security policy for details.

Check it yourself

Boris is open source under Apache 2.0, so every claim on this page can be verified in the code. The voice pipeline, agent tools, and network calls all live in one public repository.

Download Boris for Windows