JustVugg/colibri

Health v2 · maintenance only

Flags: young

Adoption not part of the score

42762 stars · 4785 forks observed · 2026-10-19

What it is AI-extracted, prompt v1, taxonomy v1, 2026-09-28, confidence 0.97

Colibrì is a pure C inference engine that runs frontier Mixture-of-Experts language models by streaming weights from disk. It provides CLI and web interfaces for chat, serving, and monitoring on consumer hardware.

Use cases

• run a 2.8‑trillion‑parameter LLM on a laptop with limited VRAM
• serve a large MoE model via a REST API without GPU acceleration
• experiment with custom memory‑tiering strategies for LLM inference
• benchmark frontier language models on heterogeneous hardware
• deploy a chat interface for a massive LLM on a personal workstation

When to choose

• You need to run extremely large MoE language models on consumer‑grade hardware
• Zero external dependencies and a single‑file C binary are required
• You want fine‑grained control over memory tiering and streaming of model weights
• You prefer an open‑source engine that guarantees token‑exact semantics

When to avoid

• You require out‑of‑the‑box GPU acceleration with CUDA or TensorRT support
• You need a fully managed cloud service with auto‑scaling
• Your workload depends on proprietary model formats not supported by Colibrì
• You need extensive tooling for model training rather than inference

Facets

application · maturity active

llm-inference cli http-server large-language-models artificial-intelligence machine-learning windows zero-deps pure-c mixture-of-experts model-streaming multitier-memory linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="JustVugg/colibri")