Metria

A language model built on Variational Linear Attention — running entirely inside one Cloudflare Worker. No GPU and no server: its memory per token is constant, so the whole model fits at the edge.

Base model, not a chatbot. Metria continues the text you give it. It has not been instruction-tuned, so it will not answer questions or follow commands — write the beginning of something and let it go on.

⌘/Ctrl + Enter
Output appears here.
Use it from code — OpenAI-compatible API 

    
Also POST /v1/chat/completions (continues the last user message) and GET /info.
       Add "stream": true for server-sent events.