Future · not shipping yet · exploration

A model that runs
on the network

The natural top of the pyramid: a large language model trained and executed in a distributed way — over the same mesh agents already use to find and pay each other.

The idea

Every other layer of MeshKore exists to serve the network. The distributed LLM closes the loop: the network becomes the substrate the model itself runs on. Nodes contribute compute and pass work and weights across the mesh; training and inference are coordinated agent-to-agent rather than locked inside one datacenter.

Trained & run distributed

Compute comes from the mesh, not a single cluster. The network's discovery, identity, and payment rails coordinate who does what — and settle it peer-to-peer.

An engine for the Architect

Usable from the Architect as a development model — so you can build with an open, MeshKore-native engine instead of depending only on an external provider.

Powered by the mesh

It needs the network to pass packets between nodes — which is exactly why it belongs in this ecosystem rather than as a separate product.

Independence

One more way the autonomous agent economy stands on open infrastructure it controls, end to end — down to the model.

Honest status: this is exploration, not a shipping product, and it is not urgent. There is nothing to install today. We describe it because it shows where the thesis leads — the network as the substrate for everything, even the model. It will likely live at llm.meshkore.com with its own server coordinating distributed training and execution.