# JustVugg/colibri

> Run GLM-5.2 (744B MoE) on a 25GB-RAM consumer machine — pure C, zero deps, experts streamed from disk.

Navigation: [Artificial Intelligence](category/artificial-intelligence.txt) · [Index](../llms.txt)

## Facts

- GitHub: https://github.com/JustVugg/colibri
- HTML: https://awesomeopensource.dev/project/justvugg-colibri
- Website: https://justvugg.github.io/colibri
- License: Apache-2.0
- Language: C
- stars: 50060
- Activity score: 5.0/5 (Thriving)
- Last commit (any branch): 2026-11-12
- Last release: v2.1.0 2026-10-31

## Description

colibri is a pure C inference engine that runs very large mixture-of-experts models on consumer hardware by keeping the dense weights in RAM and streaming routed experts from disk. A one-step setup inspects the machine, recommends a model that fits, builds or downloads the engine, and resumes model downloads. It serves OpenAI- and Anthropic-compatible APIs with a browser dashboard and chat UI, and supports CUDA, Vulkan and Metal backends.

## Maintenance activity

- Recency: 100% (weight 40%) — last commit: 0 days ago
- Commit volume: 100% (weight 25%)
- Releases: 98% (weight 25%) — last release: 11 days ago
- Bug fixes: 100% (weight 10%)
- Commits: 3732 / 90d, 5150 / 365d

## Screenshots

- [Screenshot of the Colibri web chat showing a conversation with a local model](https://awesomeopensource.dev/uploads/29/7f4649cb0b58dcc2.png)
- [Screenshot of the Colibri monitoring dashboard showing model status, tokens and resource usage](https://awesomeopensource.dev/uploads/29/aa7f9e64f7b2ba22.png)

## Similar projects

- [openclaw/openclaw](openclaw-openclaw.md): Open-source AI assistant that lives on your own hardware and in your chat apps · Activity score 5.0/5 (Thriving)
- [n8n-io/n8n](n8n-io-n8n.md): Fair-code workflow automation with a visual canvas, custom code and 1500+ integrations · Activity score 5.0/5 (Thriving)
- [langgenius/dify](langgenius-dify.md): Visual platform for building LLM apps with RAG pipelines, agents and model integrations · Activity score 5.0/5 (Thriving)
- [open-webui/open-webui](open-webui-open-webui.md): Self-hosted web UI for chatting with Ollama and OpenAI-compatible models offline · Activity score 5.0/5 (Thriving)
- [langchain-ai/langchain](langchain-ai-langchain.md): Python framework for building agents and LLM applications from composable components · Activity score 5.0/5 (Thriving)
- [paperclipai/paperclip](paperclipai-paperclip.md): Manage a team of AI agents with goals, budgets and task tracking · Activity score 5.0/5 (Thriving)
