metadata
title: README
emoji: 馃獝
colorFrom: red
colorTo: gray
sdk: static
pinned: false
govai labs
Small, fast, private AI that runs on your own computer.
Clear answers in plain words 路 free and open 路 no account, no cloud, no tracking
What we make
| Product | What it is |
|---|---|
| 馃 govai 2.3 | A ~2 billion parameter chat assistant in one 1.3 GB file (GGUF, Q4_K_M). Plain-language answers, step-by-step guides on request, maths and science reasoning, and an optional thinking mode. Runs with Ollama, llama.cpp or the govai app. |
| 馃捇 govai for Windows | The desktop app: chat with govai fully on your PC, web search with sources when a question needs fresh facts, a built-in calculator, memory of what you tell it, light and dark themes, and updates inside the app. |
| 馃摫 govai for phones | Android and iPhone app, in development. |
Why govai
- Private by design. The model runs on your own device. Your chats and memory stay there.
- Answers that fit the question. A line or two for quick questions; a full guide with headings, examples and an "In short" summary when you ask to "explain fully".
- Strong for its size. 81% on school maths word problems (GSM8K), 82% on science reasoning (ARC-Challenge), 59% on competition maths (MATH-500).
- Good everyday answers. 9.0 / 10 on our test of real everyday questions, judged for being correct, clear, complete and well formatted.
- Fast on ordinary hardware. About 150 tokens per second on a gaming GPU, and usable on a CPU.
- Your language. English first, plus Hindi and Telugu.
Try it in one line
ollama run hf.co/govailabs/govai:Q4_K_M
Or with llama.cpp: llama-cli -hf govailabs/govai --jinja. Settings, benchmarks and limitations are on the model card.
govai labs 路 open weights under Apache 2.0 路 built for people who want a helpful AI they control