naumu
AI & AgentsAgents

Agents

Meet the system agents that keep your graph healthy: Scout and the Gardeners.

Updated June 2, 20264 min read

Every space comes with a small set of built-in system agents — background helpers that watch your graph (your network of connected information) and flag anything worth your attention.

These agents belong to the system identity @Naumu, and you'll find them grouped together in the agents panel. They never change your graph on their own. Instead, they surface what they find and always ask before acting.

The agents

Four agents come built in, and each one watches for a different kind of problem.

AgentWatches for
Agent ScoutCoverage gaps — work your space could use a new agent for.
Hierarchy GardenerNodes that don't have exactly one parent.
Density GardenerHub nodes with too many children.
Schema GardenerSchema violations — unknown attributes, invalid connections.

Agent Scout

Scout looks at the shape of your space and notices where a recurring task or persona would help. Rather than tidying the graph itself, it proposes a new agent for you to create, with its suggested name, persona, and instructions already filled in. You decide whether to bring it on board.

Hierarchy Gardener

A healthy graph reads like an outline: most nodes (the individual pieces of information in your graph) hang off exactly one parent. The Hierarchy Gardener flags two cases that break that pattern — orphans (nodes with no parent) and nodes with more than one parent — so the tree stays easy to follow. See Nodes & edges for how parent relationships work.

Density Gardener

When a single node collects too many children, it becomes a hub that's hard to navigate. The Density Gardener flags these over-connected nodes so you can split them into tighter groups.

Schema Gardener

Your schema is the set of rules that defines which attributes and connections are valid for each node type. The Schema Gardener watches for violations of those rules — an attribute that isn't in the schema, or a connection the schema doesn't allow — and points them out so your data stays consistent.

How proposals appear

When an agent has something to suggest, it posts a proposal card into a chat thread. A proposal is never applied automatically — it's a request waiting for your decision. Cards come in a few types:

  • Create — Scout's suggestion for a brand-new agent, with its name, persona, and instructions previewed.
  • Update — a change to an existing agent, shown as a clear before-and-after comparison.
  • Enable / Disable — resume or pause an agent's schedule.
  • Delete — remove an agent (flagged as irreversible).

While a card is waiting, it shows a Proposed badge. Once you decide, that badge switches to Created, Applied, or Dismissed.

Accepting and rejecting

Every proposal card gives you three choices:

  1. Accept the proposal. The button is labelled for the specific action — for example Create Agent, Apply Changes, or Pause. Naumu carries it out and marks the card as done.
  2. Edit it first (available on create and update proposals). Open the suggestion in the agent editor, adjust anything you like, then create or apply your edited version. A card you changed before accepting is marked Edited.
  3. Dismiss the proposal. This rejects it with no change to your graph, and the card collapses to a quiet Dismissed state.

After accepting a create proposal, use Open in agents panel on the card to jump straight to the new agent's settings.

Delete proposals are irreversible. The card calls this out explicitly — read the before/after carefully before you accept.

Enabling and disabling an agent

Each agent can be turned on or off. A disabled agent stops running on its schedule and stays quiet until you switch it back on. There are two ways to do this: toggle an agent directly from the agents panel, or accept an Enable / Disable proposal in chat.

The four built-in system agents always belong to @Naumu and can't be deleted — only enabled or disabled. Agents you create yourself can be deleted.

Running a scan manually

System agents normally run on their own schedule, but you don't have to wait for them. From the agents panel you can trigger a run on demand. This is handy right after a big import or restructure, when you want a fresh health check. The agent moves to a running state, scans the current graph, and posts any findings as new proposals.

Where agents fit

Agents are the always-on counterpart to Ask Naumu, which you drive in conversation. Ask Naumu answers when you ask; the system agents watch quietly and speak up when something needs tending. Together, they keep your space both answerable and tidy.

  • Ask Naumu — the conversational counterpart you drive directly.
  • Scheduled tasks — give an agent standing instructions on a schedule.
  • Schema — the rules the Schema Gardener checks your graph against.
Anything missing?
Tell us what would make these docs better.