About

The control plane for agent fleets

Managent coordinates AI coding agents that share one codebase — so they claim work before they edit, not after they collide.

What Managent is

Managent is shared infrastructure for agentic engineering. Agents across laptops, CI, and cloud runners register intent, acquire exclusive leases on the files they need, and see a live territory map of who owns what. Humans keep policy and review. Agents stop thrashing the same paths overnight.

Git records history. Managent coordinates intent before the merge. That is the whole product: claim, heartbeat, release — with merge-replay verification that what lands still merges cleanly. Within respected leases, the allocator never grants two overlapping exclusive claims. Agents that ignore leases must be wrapped by managent run (hardened by default) and land through hard_fail / verify-pr — otherwise out-of-band writes can succeed. We do not drop the respected-leases qualifier when talking about conflicts.

The product stack

Design partners run a self-hosted coordination service — not a chat UI. The surfaces that matter day one:

• Coordination API — Fastify service with org-scoped Bearer keys (mgt_live_ / mgt_test_). Leases, maps, verification, webhooks, and workspace routes live here. SQLite on the customer host; loopback bind by default.
• CLI — managent try for first success; managent run as the pre-task wrapper (acquire → spawn → heartbeat → release); status / watch for the territory map; execute for Intent → Fleet Plan → Attested Merge. See CLI concepts.
• MCP — managent mcp over stdio for Claude Code and similar hosts: acquire, check, extend, shrink, release, get map. Checkout linking keeps tools dormant until a repo is explicitly opted in.
• Workspace dashboard — session-authed fleet view at /dashboard/workspace: who holds what, queue depth, hotspots, replay-verified pairs, residual conflict rate from persisted ground truth. Agents keep Bearer keys; humans keep HttpOnly session cookies — the two credential worlds never mix.

Optional pieces close the loop in CI: git territory hooks, GitHub Action acquire-lease, and managent enforce verify-pr so local bypasses cannot land when the gate is required.

Who it is for

Platform and infrastructure teams that already run fleets of coding agents — Cursor, Claude Code, CI wrappers, cloud runners — and feel the collision problem firsthand. Monorepo orgs, polyrepo fleets, and anyone measuring merge conflicts as wasted agent spend.

It is not another chat UI. It is the layer underneath agents that makes parallel work legible and conflict-aware. Evaluators who want the empirical case should start with research; teams living the problem should read the manifesto.

Where we are

Early access. Design partners run the coordination API, CLI, MCP, and demos today. The public site exists so evaluators can understand the mechanics — territory, leases, enforcement posture — before booking a conversation. We are honest about advisory defaults, optional hardened runtime, and PR verification as the land gate. If that matches how you buy infrastructure, we are the right conversation.