# Hyverk pool > An open-weights LLM served by machines that lend part of their memory. A > hub at https://llm.meshkore.com plans which model fits, splits it across > the lending machines with llama.cpp, and relays their traffic. Each > machine runs `hyverk-node`, which also gives its owner an > OpenAI-compatible endpoint on 127.0.0.1:17080. No accounts and no forms: > a node is an Ed25519 key it creates on first run. Every page below is Markdown and can be read with one GET. An agent can set up, run and use a node without a human filling anything in. ## Docs - [Overview](https://llm.meshkore.com/docs/index.md): what the pool is and how a request travels - [Join](https://llm.meshkore.com/docs/join.md): the desktop app, or the pinned llama.cpp release plus `hyverk-node pool`, step by step for agents - [Use](https://llm.meshkore.com/docs/use.md): call the model through the local endpoint - [Security](https://llm.meshkore.com/docs/security.md): what is protected, how, and the known gaps - [Credits](https://llm.meshkore.com/docs/credits.md): what lending will earn, and what is not built yet - [Protocol](https://llm.meshkore.com/docs/protocol.md): hub endpoints, signatures, links ## Live data - [Stats](https://llm.meshkore.com/pool/v1/stats): JSON with model, nodes, tokens per second, requests, capacity - [Events](https://llm.meshkore.com/pool/v1/events): the same JSON as server-sent events, every 2 s - [Status](https://llm.meshkore.com/pool/v1/status): short form ## Source - [Repository](https://github.com/capitaharlock/hyverk-distributed-llm): `client/node` (the node), `client/tray` (the desktop app), `server/pool-hub` (the hub and this site)