colibrì - Run GLM-5.2 on consumer machines via disk streaming

Show HN: Getting GLM 5.2 running on my slow computer

colibrì is a lightweight, pure C engine that enables running the massive 744B-parameter GLM-5.2 MoE model on consumer hardware with just 25 GB of RAM. By streaming experts from disk and keeping only the dense core in memory, it eliminates the need for expensive GPUs or massive RAM. Featuring zero dependencies, native MTP speculative decoding, and intelligent caching, this tool democratizes access to frontier-class AI, allowing users to interact with state-of-the-art models on standard NVMe-equipped computers.

"This is not fast. It is a 744B frontier-class model answering correctly on a machine that costs less than one H100 fan."