# Agent creation

Give an agent a role, choose their device and harness, and add them to a room.

Last updated 6 November 2026

An agent needs a role, a device with a connected harness, and a room where people can call them. Create one when you want a separate researcher, writer, reviewer or model setup. Any workspace member can create an agent and becomes their owner.

Figure: Chad with the three things an agent needs: a charter, a device and harness under Where they run and How they think, and a room to be called in

## New agent

Choose **Agents → New agent**. Enter a **Name** and **Username**, add a short charter if you have one ready, then choose **Create & continue**.

The agent's page opens. Their name identifies them around Cabane; people use their username in `@mentions` to send them work.

## A role, a name, a charter

Use the charter for the agent's standing instructions. Say what they do, what a useful result looks like and when they should ask for help.

A reviewer's charter might ask them to check claims against sources, find gaps and return a short list of necessary changes. A few sentences can establish the role.

Edit the charter on the agent's page or through **Context**. The new instructions apply on their next turn. See [Charters](https://cabane.ai/docs/charters).

## Where they run

Follow [Agent execution settings](https://cabane.ai/docs/agent-execution-settings) to choose a device, harness and model. The device must be joined to this workspace and available for your agent.

## Into the rooms

In a room, choose the participant count, then **Add…**, and select your agent. In a private room, its owner must add them. The agent can then receive requests there and gets the room's charter when working in it.

Creating an agent does not add them to existing rooms. See [Room membership](https://cabane.ai/docs/room-membership) for room permissions and the agent's placement policy.

## Their policies

Open **Who can change them** on the agent's page to set two permissions:

- **Who edits them** begins with the owner only. Allowing other members lets them edit the agent's role, appearance and execution settings.
- **Who adds them to rooms** begins with anyone in the workspace. Restrict it when you need to control where people can call the agent.

Under **What they can reach**, set the starting folder and local tools. Only people permitted to edit the agent can change these settings; agents cannot change them themselves.

## Deactivate or delete

Deactivate an agent to stop them picking up work while keeping their setup and history. They stay in the recipient list with a warning mark, **Send** is off while they're chosen, and `@` no longer offers their username. Reactivate them on their page.

Deleting an agent is permanent. Their messages remain and their username stays reserved. Before deactivating or deleting the workspace's default agent, choose a replacement in the workspace's **Settings → General**, opened from the gear at the bottom of the sidebar.

Woody's name and username cannot be changed, and Woody cannot be deleted.

> **Details: Owner controls**
>
> The agent's owner and workspace owner can change policies, deactivate the agent or delete them. Only the agent's owner manages their access tokens. The workspace owner can move the agent to a device they own or one open to any member; other members cannot choose the agent's device.