AsistAI hears you, understands and answers with a voice - like a real conversation partner. All neural networks run right on your PC: no clouds, no internet, no subscriptions. Your conversations stay with you.
AsistAI is a voice assistant that lives entirely on your computer. Say its name - it will hear you. Ask a question - it will answer with a voice. No clouds and no foreign servers: speech recognition, the "mind" and the voice all work locally, and computations can run on your graphics card.
Nothing to install: the archive contains a single ready-to-run file asist.exe. Launch it - and you are already talking.
Six steps - from your voice to a living answer. Everything happens on your computer.
An adaptive echo canceller subtracts the assistant's own voice from the microphone signal - it does not "hear" itself.
The assistant listens for its name. Once it hears it, it starts recording speech and stops at a pause of silence.
Your speech is turned into text by a local Whisper model - on your own computer.
A local language model thinks through the answer. It knows how to use tools: volume, launching apps, time and date.
For Russian - normalization, stress placement and numbers written out. The answer sounds natural, like a person.
Text is turned into speech and plays from the speakers. Long answers are voiced in parts.
Everything you need for a comfortable conversation with the assistant - and nothing extra.
No subscriptions, keys or "paid features". Download it - use it as much as you like.
All data stays on your computer. Not a single word goes to the internet.
Full autonomy: it works on a plane, in a basement, and anywhere there is no signal.
Minimal resource usage. The LLM can be accelerated by your graphics card (Vulkan).
Stress, abbreviations and numbers written out - the voice sounds natural, like a person.
Russian, English, German, Spanish, French and Hindi.
Name the assistant however you like - and address it by that name.
You can swap the AI model that forms answers - to fit your tasks and taste.
A female or male voice - choose whichever feels more familiar.
When needed, the assistant fetches up-to-date data from the internet.
No Python, third-party modules or shells needed. One file - and everything works.
AsistAI is designed for modern computers. Here are the recommended specs for a comfortable experience.
The AVX instruction set speeds up neural network computations.
An NVMe SSD for fast model loading and a fast start.
Enough to run all of the assistant's components in parallel.
Hardware acceleration of the LLM via the Vulkan API.
So the assistant can hear you clearly.
So the assistant can answer with a voice.
These are recommendations, not strict requirements: on a weaker computer the assistant will still start, it will just think a little longer.
A voice assistant that works fully locally and autonomously: it listens to the microphone, recognizes speech, forms an answer with a local language model, places stress, and synthesizes and voices the answer. No cloud APIs - all models run locally, and AI inference can be performed on a GPU. No cloud, no subscriptions and no internet services.
This project is a side project; the main goal is to create an AI portable wearable autonomous assistant and an assistant for controlling smart homes and corporate infrastructure.
At present, all similar developments are dependent on third-party organizations and services, which is extremely undesirable - I want to have my own autonomous, controllable software.
This assistant works very poorly with radio/Bluetooth/Wi-Fi speakers and microphones - it will work, BUT the echo cannot be filtered, which is a serious problem for projects like this.
There are two solutions: