colibri
by JustVugg
Colibri is a small program written in C that runs very large open AI language models on an ordinary computer by keeping only the active parts in memory and reading the rest from disk. It is for people who want to chat with big models, or serve them to other apps, without a high-end GPU or a cloud subscription.
- Stars
- 45.4k
- Licence
- Apache-2.0
- Runs on
- Your computer
- Setup
- Moderate
- Language
- C
- Age
- 122 days
What you can do with it
- Chat with a large open language model on your own computer from the command line or a browser dashboard
- Point other apps at its local OpenAI-style or Anthropic-style address so they use your own model
- Run existing GGUF model files on the CPU, or on a GPU through Vulkan or CUDA
Make money with it
Set up private, locally hosted AI assistants for small firms that cannot send their data to cloud services, and charge for installation and support.
An idea, not financial advice. Check the licence before charging for anything built on it.
Ways a business could use it
- Run a private AI assistant for drafting and summarising documents without sending data to a paid cloud service
- Offer a local OpenAI-compatible endpoint to internal tools so you avoid per-token API fees
Can you use it commercially?
Free to use, change and sell, including in paid products. Keep the copyright notice.
Not legal advice. Read the project's licence file before building a product on it.
Getting it running
Download or clone the repository and run START-HERE.bat on Windows or ./start-here.sh on Linux and macOS, which checks your machine, builds the engine and downloads a suitable model.
Follow the official instructions in the project README.
Things to check
- Needs at least 8 GB of RAM and 22 GB or more of free disk space for the smallest model
- Streaming experts from disk will be slow compared with a GPU or cloud service
- Model downloads are large and each model has its own licence to check
Who's talking about it
Checked 01 Nov 2026. Summary drafted by AI from the project's README.
