Enough to get from nothing to a published site on a server you own. This is a getting-started guide, not a reference manual — the reference does not exist yet, and the last section on this page says exactly what is missing.
To build: a Mac, a Linux box or a Windows machine. That is the whole list. The first project runs entirely on it — no account, no server, no network.
To publish somewhere public: a server you rent or own with Docker on it, and a domain if you want a real address. A small VPS is enough.
Mother is the desktop app you build in. Download it for your platform and open it — it starts its own Host in the background, with a Postgres, a runtime and object storage all inside the app. There is nothing to configure and no account to create before you can build.
The app updates itself in place. In App settings you can let it fetch updates in the background and install them the next time you launch.
Create a project and you get a page tree, an empty Data section and a canvas. A useful first pass, in order: add a collection and give it a couple of fields; add a page; drop a collection list on it and bind it to that collection; design one item, and every record follows the same design.
Styling goes on named classes rather than on the individual block, so the second card costs nothing. Edits save themselves. Preview opens the page exactly as a visitor would get it, without publishing anything.
On the server, fetch the CLI, install the stack, and attach the machine to your account:
install runs pre-flight checks on Docker, ports, disk and connectivity, pulls the pinned image, writes brood.yaml, .env and docker-compose.yml into ~/brood-host, starts the containers and waits until the Host reports healthy. Running it again is safe — existing passwords and keys are kept.
By default the install is self-contained: it brings its own web edge and object storage, and --edge-domain turns on automatic HTTPS for that name once DNS points at the box. If the Host is going to sit behind a proxy you already run, use --lite instead and terminate TLS yourself.
pair registers the machine with your account, so Mother can see it and publish to it over the tunnel. The machine's secret is minted on the machine and is never stored on our side.
The same CLI operates the Host afterwards. Nothing here needs our servers to be up.
status prints containers, database and tunnel health on one screen. update asks which image version is currently recommended, compares it with the pin in your brood.yaml, and rolls the stack over only if there is a newer one — you choose when. doctor writes a timestamped diagnostic bundle to hand over when something is wrong.
There is also brood self-update for the CLI itself, and brood uninstall to take the stack back off the machine.
Brood has no AI builder of its own. It exposes an MCP endpoint instead, and you point your own agent at it — Claude, Claude Code, Cursor, anything that speaks the protocol. From there the agent works on the real project: pages, collections, workflows, media, publishing. Not a sandbox and not a code generator writing files somewhere else.
You bring the agent and the key; we do not resell inference. A free account gets 50 operations, which is enough to decide whether you want it. This very site is built and edited that way.
Eject compiles the project into a standalone repository: Next.js components generated from the block trees, a Drizzle schema generated from the real database, the migrations, your data, and a Docker Compose file to run the whole thing. There are no Brood packages in the output and nothing in it phones home.
It is free on every plan, and it is tested on every push to Brood itself — a freshly exported project has to install and build, so "it exports" cannot quietly stop meaning "it runs". Start the export from the project's own page in Mother.
Some interactions are still catching up on the exported side. The features page marks that honestly rather than hiding it.
So you do not go looking for pages that are not there:
Fixes and new sections show up on the changelog.