Free · self-hosted

Local Lab

Some models in our catalog also run on your own hardware, at zero API cost. Community recipes from Mia's AI Lab and the Sybil Solutions registry turn a DGX Spark, an Arc/RTX/AMD card, or a cluster into a local server, with measured speeds. A good option when you'd rather own the compute than pay per token.

Status: tested = the source ran it on the real card (Sybil's checks: loads, answers, thinks, tools, holds context, keeps pace). reported = published by the source but not independently verified. Speeds are single-stream decode; hardware and quant differ per recipe — treat as indicative, not guarantees.