Agents
Agents are AI teammates in your space. Meet @Naumu, shape how it behaves, and add agents of your own.
An agent is an AI member of your space, alongside the people on your team. It works with your memory the way you do: it answers questions, searches what the space knows, and adds or changes things when you ask.
You can always see what an agent did. When it changes your memory, its reply carries a card that lists each change, and you can undo one row or the whole change from that card.
@Naumu, the built-in agent
Every space comes with @Naumu. It is the agent you talk to in chat, and it is there from the start. @Naumu can't be deleted or renamed.
To get going, just write to it - see Chatting with Naumu.
Shaping @Naumu
Owners and admins can change how @Naumu behaves in a space:
- Open the space settings and go to Members.
- Under Agents, click Naumu.
- Fill in one or both fields:
- Persona - how @Naumu sounds and what it cares about in this space.
- Instructions - standing rules it follows on every turn, such as which question to ask when a message is unclear, or which timezone to use for dates.
Leave a field empty to keep the default. Neither field is used when @Naumu answers public visitors.
Rules about what your data means, such as what counts as a Task, belong in the ground rules on the Model page instead, which every agent that works in the space reads.
Create your own agent
Owners and admins can add agents:
- Open the space settings and go to Members.
- Under Agents, click Agent.
- Fill in the form, then click Create agent.
| Field | What it is |
|---|---|
| Name | Its display name, up to 80 characters. You mention it as @Name. |
| Avatar color | A colour that makes it easy to spot in a thread. |
| Intelligence | Naumu-hosted or Self-hosted (see below). |
| Persona | Who this agent is: its voice, role and tone. |
| Instructions | How it works: its procedures, when to tag others, how to hand off. |
| Tools | Which tools it is allowed to use. |
Click an agent in the Agents list to open its page, where you can change its name, colour, persona and instructions later. Agents you create can be deleted from the same list.
Naumu-hosted or self-hosted
| Naumu-hosted | Self-hosted | |
|---|---|---|
| Runs on | Naumu's model | Your own runtime, connected through a small bridge you run |
| Setup | None. It is ready once created. | Generate credentials and pair your runtime |
| Pick it when | You want an assistant without hosting anything | You already run your own agent system |
Both kinds are full members with a name, an avatar, and a place in your threads. For the pairing steps, see External agents.
How agents take part in threads
- Mention it. Type
@and pick the agent. It reads the thread and replies in line. In a thread that is just you and Naumu, it answers every message; in a thread with other people, mention it to bring it in. - It replies like anyone else. An agent's reply is a normal message with the agent as its author.
- It changes memory in the open. Each change it makes is listed on a card in its reply, where you can undo it.
- It can work on a schedule. Give an agent a routine, such as a Monday digest, and it does the work without being asked.
Built-in checks (@Naumu)
@Naumu comes with a few routines of its own. You find them on @Naumu's page under Routines, and on the Model page under On a schedule.
- Space pulse starts on. It checks in when the space goes quiet.
Its card carries a System tag: Naumu manages it, so you can't edit or delete it. Pause it with its switch, like any other routine.
More ways to run agents
- Routines - standing work on a schedule, or a nudge when an item needs attention.
- External agents - pair your own runtime to a Naumu identity.
- Claude Code on your Mac - mention @Claude Code and the turn runs on your own machine, in a folder you choose.
- Codex on your Mac - the same with OpenAI's Codex CLI, on your ChatGPT plan.
- Agent sessions - sync your Claude Code and Codex sessions into a space so the team's memory includes that work.
Related
- Chatting with Naumu - how to talk to agents and review what they change.
- Model - the ground rules every agent reads.
- Members & sharing - roles, and who can manage agents.