Quick start

Notarium is a self-hosted knowledge base built as plain Markdown files on your own disk — one that a human and an AI agent work on as equals. No cloud, no proprietary format: notes live in folders, and git can be the source of truth.

This section is the shortest path from zero to a running instance: deploy Notarium in a single container, create your first note in the web editor, and connect your agent over MCP. The whole thing takes about five minutes.

What you'll need

Just Docker (or Docker Desktop) and a free port. Everything else — the web editor, the engine, and the MCP endpoint for agents — is already inside the image. There's no need to install Node, a database, or a vector index separately.

How it works

The same knowledge base is open to two readers at once. The human works through the web editor, the agent through the MCP endpoint; both see the same note graph, and on disk it's just a tree of .md files.

flowchart LR
  human([Human]) -->|web editor| core
  agent([AI agent]) -->|MCP · start_session| core
  core[Notarium engine] --> files[(Markdown files<br/>on disk)]
  files -.->|git = source of truth| git[(Repository)]

The key principle is file-first: a note is a file, not a row in someone else's database. There's no need to export, because you're never really "inside" — the data is always yours and always on disk.

Spin up an instance

The fastest way is to run the prebuilt image. It's self-contained: it brings up the engine, the web interface, and the MCP endpoint all at once.

# Pull the image and start the container
docker run -d --name notarium \
  -p 3000:3000 \
  -v notarium-data:/data \
  docouno/notarium:latest

A single /data volume holds all state: the metadata database, the indexes, your notes, and export artifacts. There's nothing more to configure — port 3000 and the data path /data are already baked into the image. The port to the left of 3000 can be changed to any free one.

Open http://localhost:3000 — you'll be greeted by the first-run setup screen. It creates the instance owner (the first account) and a starter (personal) space. After that, you're in the editor.

Port and data

You can change port 3000 by editing the left-hand number in -p <yours>:3000. The docouno/notarium image is public — this is the main install path (details and building from source are on the Installation page). The /data volume survives when you recreate the container — but guard the volume itself: your data leaves with it.

What happens under the hood on first launch:

the engine starts and serves the web interface on the port you specified;
the tree of your spaces is initialized inside the /data volume;
the setup screen creates the owner — the account that owns the instance.

Next steps

The instance is up — three short steps come next. Each is covered on its own page:

Step	Page	What you'll get
1	Installation and launch	A walkthrough of `docker` / `make up`, volumes, and the setup screen
2	Your first note	The file tree, the web editor, saving to `.md`
3	Connect an agent	A token, the `POST /mcp` endpoint, calling `start_session`

Want to grasp the whole model first? Take a look at the Concepts section: spaces and projects, note types, the knowledge graph, and the access model for humans and agents.