novamem · docs

Appearance

Projects (sub-brains) ​

A project is a shared scope for memory. Use them when you want a body of memory visible to a group of people instead of just yourself.

Mental model ​

Every entry has exactly one owner-user. By default that's you and it's invisible to everyone else. Adding the entry under a project widens the visibility to every member of that project — and only those members.

user ─┬── private memory                          (only you see it)
      │
      └── project memberships
            ├── alice's project   (alice + bob + you all see entries here)
            └── platform team     (everyone in `platform-team`)

Memory crosses user boundaries only through explicit project membership. There is no "share a single entry"; sharing is at the project granularity.

Create ​

Projects page → Create project → pick a human name. The id is a server-assigned ULID. You become the owner automatically.

bash
curl -X POST https://novamem.example.com/v1/me/projects \
  -H "Authorization: Bearer ns_..." \
  -d '{ "name": "Phoenix" }'

Activate (sidebar switcher) ​

The sidebar's project switcher sets the active scope for subsequent search / recent / today / graph calls. The switcher persists to localStorage (nm-active-project) so it survives page reloads.

  • Global memory only — searches your private user-global store.
  • A specific project — unions user-global with that project's entries.

Activating doesn't write anything; memory_remember calls still write to user-global by default unless you pass project: "<name|id>" explicitly.

Share ​

Project page → expand a project → Add member → username. The member can read + write the project's memory immediately. Removing a member revokes access immediately too — no re-encryption needed because there's no per-user encryption layer (security boundary is the database, not crypto).

Roles ​

Within a project there are two roles:

OwnerMember
Read project memory✓✓
Write project memory✓✓
Add / remove other members✓✗
Delete the project✓✗ (can leave only)

The owner is the user who created the project. Ownership doesn't transfer; if the owner leaves, the project goes with them via the Delete project action (which cascades all entries + cold collections + graph nodes).

Delete vs leave ​

  • Delete (owner only) — full cascade: every memory entry across every namespace, all cold-tier Qdrant collections, all graph nodes. Irreversible.
  • Leave (member only) — drops your project_members row. Project keeps existing for the other members. You no longer see its entries.

Why projects, not tags? ​

Tags are easy to forget; projects are explicit. The visibility boundary is the same primitive that gates every search and every write — there's no path where a private memory leaks via a tag-like field. If you want intra-project organisation, use namespaces within the project.

See also ​