Federation

Brains can form a hierarchy: a person's brain inside a team's, a project's inside a company's. Each brain declares where it sits and what it is willing to share upward. Sharing is closed by default: a brain that declares nothing shares nothing.

Where a brain sits

scope in client.config.json places the brain on the scope ladder (federation, company, domain, project, workspace, person, or agent for an autonomous node):

client.config.json (excerpt)

{
  "scope": {
    "level": "project",
    "id": "checkout",
    "parent": "company:example",
    "peers": []
  }
}

Ownership flows down, summaries flow up, and access narrows on the way up.

What a brain shares

federation is the publish-upward contract. publishes lists the kind ids a parent may receive; everything else is withheld.

client.config.json (excerpt)

{
  "federation": {
    "publishes": ["doc", "wiki-page"],
    "withholds": [],
    "children": [],
    "enrollments": []
  }
}

After make build, app/brain-manifest.json states the contract under policy:

{
  "policy": {
    "publishes": ["doc", "wiki-page"],
    "withholds": ["action-item", "decision", "memory-note", "person", "..."]
  }
}

Every declared kind that is not published appears in withholds, so the manifest always says plainly what stays home. Kinds you list that your ontology does not declare are ignored.

Serving across brains

The engine also contains a federation service that answers searches and context requests across several brains at request time, rechecking each source brain's own read policy on every call. Running it needs a separate authenticated host for each realm and operator configuration that this release does not package or document yet. For now, federation in this release means the declarations above.