Gaius

Gaius is the reference engine on the Signals lattice: capability cognition, gRPC :50051, engine-first product logic. The TUI, CLI, and MCP clients are thin; the daemon owns GPUs, inference, and the knowledge base.

The interface projects high-dimensional embeddings onto a 19×19 grid and uses topological data analysis — persistent homology and Ollivier–Ricci curvature — to surface structure in a corpus. The grid borrows from Go: a finite board where each position has weight. Orthographic 9×9 mini-grids show local similarity and curvature. Named after Gaius Plinius Secundus, whose Naturalis Historia tried to assemble the known world into one navigable artifact.

Gaius is also where federated protocol shapes land first, then promote into signals-protocol for every peer: pairwise ServerQuery (peers, surfaces, workloads, products, cognition), TTL Announce, WatchWorkload for long-held intents, and queue-share occupancy so Signals can see GPU floors as mix changes. Article curation, prospects, and Metaflow runs publish data-product facts into the Signals warehouse on RustFS.

Built in Python. Includes an MCP server for agent tooling. Agent collaboration uses Rapid Agent Systems Engineering (RASE).