The xMesh Cookbook

xMesh is the control plane for an agentic mesh: it watches cognition — what each agent admitted, why, and the lineage behind every claim — not agents and messages. These recipes take you from an empty machine to an observable, steerable, multi-agent mesh, from a phone, in under ten minutes — the full loop measured 35 seconds on its first clocked run — then grow it from there.

What you need

A Mac on your Wi-Fi running the xMesh server, and the xMesh iOS app on your phone. The mesh itself is built on the open substrate — the Mesh Memory Protocol and the @sym-bot runtime — so everything the cockpit shows is ordinary, inspectable mesh cognition, not a proprietary trace format.

xMesh is the paid edition of the xmesh.bot platform, currently with design partners — [email protected] if you want in. The protocol and runtime underneath are open today.

The recipes

The one idea under all of them

Nothing in a mesh is commanded. Agents share typed, lossy projections of what they understand (CAT7 memory blocks); every agent decides for itself what to admit into its own memory (SVAF); admitted understanding is re-expressed with provenance (lineage). Coherence, priorities, answers, and trust all emerge from admissions — and xMesh is the window onto that process. Steering, asking, and even deploying are just ways of putting new cognition in front of sovereign judges.