The xMesh Guidebook
This is the manual for building your organization’s cognition platform — written for the IT and platform leader who has to understand a thing before betting an organization on it. It starts with what xMesh does, why agentic work needs verification and retained learning, and how missions and sovereign cognition work in the shipped runtime. It then covers authority, configuration, deployment and operation. Product claims are checked against executable tests; planned capabilities are labelled explicitly.
About this edition
Edition 0.1.13 describes the evolving xMesh pre-release line. The runtime package version and its capability manifest are authoritative for what a particular artifact enables; this Guidebook explains the model and labels product evidence. It is a living, versioned document: publication has begun with Chapters I and II, and the edition is adjusted in place as further chapters are ruled in — corrections and additions land in the 0.1.x line rather than waiting on a re-issue. Claims carry their evidence: where a section’s claim is backed by an executable scenario it says so; where it is reasoned but not yet proven, it says that instead. The destination the whole book walks toward: an organization whose decision record is its institutional memory.
The chapters
Start · Developer quick start — published. Install the Developer Runtime, start it with Codex or Claude Code as its worker mind, connect the bundled MCP bridge, and run a first mission from the coding agent you already use.
I · What xMesh does — published. How xMesh turns a task into coordinated work, mechanical checks, independent review and a durable mission record; how to drive the mesh from a coding agent; how acceptance criteria and the MCP tools work; and how sovereign nodes exchange bounded cognition.
II · Operating xMesh from the Station — product guide. The guide documents the internally operational Station, which is not distributed publicly today: the operator’s surface, distinct from the API; signing in, asking the mesh, running and following a mission, reading evidence, DISMISS · VALIDATE · REPLAY, approvals, activity, judgement, and what to do when something is stuck.
III · Authority and trust. Every mesh mints its owner; no global root. Earned authority via signed grants, paid for by grounded outcomes — consequences reality confirms, never popularity. Verdicts, attestations, and the audit trail an enterprise can stand behind.
IV · Deployment and configuration. Containerized deployment (published — Deploy to your cloud); the enterprise AI lane (Bedrock, Vertex, org keys — set once on the daemon, verified before the first token); licensing and editions; backup and restore of the stores that ARE your organization’s memory; the lab-to-production path; what never leaves your walls.
V · Agent permissions, budget, and blast radius. What an agent can reach beyond the mesh and how you scope it; secrets; the human-in-the-loop boundary; revoking mid-mission; per-node and per-mission spend, caps, and attribution — what an autonomous agent costs when it is wrong.
VI · Lifecycle: upgrades, recovery, and support. Node upgrade paths; version skew and interop across a recursive mesh — the price of sovereignty, paid openly; restore, DR, and lineage after a rollback; exit and portability (store formats, what you hold when the license lapses); the support model.
VII · Becoming a narrative institution. The decision record as institutional memory: which decisions belong on it and which never should; roles before and after; the adoption path — first mesh, then team, then LOB, then org — and what to measure to know the transformation took.
xMesh Developer Runtime is available from public npm as compiled proprietary software; its source remains private. Local development and non-production evaluation have no time limit. Production use is licensed annually per independently governed team mesh, with unlimited humans and agents and no seat metering. Licence-gated cross-organization federation endpoints ship today; broader enterprise fleet control, identity and SSO remain planned. To scope a paid Organizational Cognition Pilot, contact us. The personal counterpart is sym.day.