A privacy-first local voice assistant built block-by-block for Raspberry Pi & Mac.
Microphone audio NEVER leaves your device. Only approved text tool requests (like asking for the weather) may leave, and every single action is recorded in a local, transparent privacy ledger.
Speech-to-text happens entirely on-device. Your raw voice is never sent to the cloud.
Only specific, user-approved text requests (like fetching weather data) can access the internet.
Complete transparency. Review a local log of every action and tool request Trusty makes.
See how Trusty stacks up against the competition.
Big tech assistants send your raw voice audio to giant cloud servers to be processed and recorded. Trusty processes all audio on-device. No cloud accounts, no subscriptions, no surveillance.
Rhasspy, HA Assist
Most offline assistants match your voice against rigid, pre-trained intent templates. You have to speak exact phrases. They lack the conversational intelligence of modern AI.
Powered by Reasoning
Trusty uses Gemma 4 as the actual orchestrator. It uses reasoning to pick tools, synthesize arguments, and understand natural language. The LLM is the brain of the home, not just a chat layer.
Everything you need, snapping together — running locally.
The brain and orchestrator, running efficiently via llama.cpp on your local hardware.
Powered by Moonshine ONNX and Kokoro ONNX for near-instant voice recognition and natural speech synthesis.
Seamlessly controls LG webOS TVs, Roborock vacuums, and more via Home Assistant integration.
Remembers names, preferences, and locations locally without ever sending your personal data away.
A friendly, animated local web interface that gives your assistant personality and visual feedback.
Hard-blocks internet tools when you want absolute grid-down privacy and complete isolation.
Open source. Free forever. Snap the pieces together your way.