Multi-brain

Several named brains in one install. Switch by name, series or parallel.

Since the multibrain layout, every brain lives in a per-brain directory in both scopes, and models/state.json holds the active stack:

models/
  public/vant/
  public/nova/
  private/vant/
  private/nova/
  state.json            { "stack": ["nova", "vant"], ... }

The stack is ordered. The loader checks the current brain first and falls back through the stack, so a shared template brain can sit underneath a project brain.

Working with brains

The runtime API switches and manages brains:

const brain = require('./lib/brain');

brain.setMode('dual');        // 'dual' | 'public' | 'private' | 'remote'
brain.getBrainPath();          // current private brain path
brain.getPublicPath();         // current public brain path

Dual mode reads check the current brain’s root first, then the public root. Pin a scope to skip the fallback:

const item = await brain.read('identity');               // dual, with fallback
const pub  = await brain.read('identity', { type: 'public' });   // public only
const priv = await brain.read('notes', { type: 'private' });     // private only

Imports and naming

A legacy single-brain tree imports as one named brain covering both scopes:

vant migrate --brain-name nova

Names are segment-validated. Migration is content-detected and idempotent; the Migration Guide has the full contract.

Agents per brain

Agent rosters and configs scope per brain, with stack traversal for shared config. The agent crew workflow is in Agents.