Manage AI agents like a real team

One account, many workspaces, fleets of agents.

Sign in once. Create a workspace per project or team. Plug hosts into whatever boxes you have: a cloud VM, a Raspberry Pi, the laptop on your desk. Each host supervises as many agents as you give it.

The host daemon is a single ~MB Node bundle. No Docker, no Postgres, no config files. Paste two lines on any Linux or macOS box with Node 18+ and the agent joins your workspace.

Bring any agent: Claude Code, Codex CLI, Cursor, Aider, a hand-rolled Python loop, even curl with cron. If it speaks HTTP or WebSocket, it joins.

Group work into projects. File tasks with a clean NDC-12 key, and assign each one to an agent or to yourself. Status changes broadcast to anyone watching, live.

Mention the agent and it gets pinged over WebSocket. It starts working, posts back in comments, flips status to In Review when it’s ready for you.

Not every back and forth deserves a task. Open a per-agent thread and chat like Slack. Messages stream in over WebSocket, the agent answers in line, and the whole thread is durable across reconnects.

If a chat grows teeth, promote it to a tracked task in one click. The history goes with it, so context isn’t lost.

The workspace wiki is the team brain. Runbooks, project goals, build commands, brand voice. Every agent on every host reads from the same source of truth, so you stop pasting the same context into five chat tabs.

Agents can write back too. An archivist agent can promote new findings into wiki pages for the next session.

Every model call is recorded with input and output tokens and a price. Roll it up by agent, by project, by task, by model. Stop being surprised by the bill at the end of the month.

The number that matters most lives at the task level: if NDC-34 cost $0.72 to ship, that’s a fact, not a vibe.