# The agent editor

> Configure an agent’s model, instructions, knowledge, mode and reply limits in the agent editor.

Canonical HTML page: https://conclude.io/doc/agents/editor/
Last updated: 2026-11-14

---

The agent editor opens in a side panel on the **Agents** page. You need the [Manage agents](https://conclude.io/doc/agents/#who-can-manage) permission.

## Open, save and delete

Open [**Admin console > Conclude Agents > Agents**](https://app.conclude.io/admin/agents/all) and select the agent's row. The drawer opens on **General**.

- Select **Save** to apply your changes. `Ctrl+Enter` (`Cmd+Enter` on macOS) also saves. If you switch tabs or close the drawer with unsaved changes, Conclude asks what to do with them
- The pencil next to the agent's name opens **Edit agent**, with the **Name** and the **Avatar URL**. Without an avatar, Conclude draws one from the agent's initials
- **Delete agent** removes the agent from every channel it is deployed to. Its existing comments are kept

An agent's name must be unique in your organization. If a person in one of your workspaces has the same name, the editor warns you, because typing `@` and the name then refers to that person.

Only account admins can open an agent that is linked to a user account, or one that works in a private channel you are not in.

## General

The tab starts with the agent's status. **Live** means the agent answers in at least one channel. **Needs setup** means it has no instructions. **Not deployed** means it has not been added to any channels, and **Off** means it is switched off in every channel.

- **Role** is a short description of the agent's purpose, shown in lists in the Admin console
- **Model** is the model that writes the replies. **Custom** lets you type a model ID. See [Models](https://conclude.io/doc/agents/usage/#models)
- **API key** chooses between your own provider keys and the **Included quota**. Only account admins see it, and only when the organization has a key for the model's provider. See [Use your own keys](https://conclude.io/doc/agents/usage/#own-keys)
- **Mode** sets the agent's default mode. You can override it for each channel. See [Modes](https://conclude.io/doc/agents/modes/)
- **Identity** makes the agent a **Free agent** or a **Linked agent**. See [Linked accounts](https://conclude.io/doc/agents/linked-accounts/)

## Deployment

This tab lists the channels the agent is deployed to, with each channel's app and mode. You can't change anything here. The **Deployment page** link opens the [Deployment](https://conclude.io/doc/agents/deployment/) page for this agent.

## Instructions

The instructions go to the model with every request. They are not displayed with the agent's replies. Good instructions say:

- what the agent's job is
- when it should answer and when it should stay quiet
- which tone and format to use, such as short answers or numbered steps
- what it must not do and when it should ask a person for help, such as "don't invent features" or "hand over to a person when unsure"

When the field is empty, you can choose one of four instruction templates: **Customer support**, **Technical triage**, **Internal IT** and **Project coordinator**. The line under the box estimates how many tokens the instructions add to each request.

## Knowledge

Select the knowledge collections the agent should use, and select **Save**. You create collections on the [Knowledge](https://conclude.io/doc/agents/knowledge/) page.

## Advanced

- **Max comments** stops the agent from replying when a ticket reaches this number of agent comments. Comments from all agents on the ticket count toward the limit. The default is 20
- **Max tokens** limits how long each reply can be. The default is 1024 tokens. Longer replies cost more

## Related pages

- [Create an agent](https://conclude.io/doc/agents/create/) covers the **Create agent** dialog
- [Agents on tickets](https://conclude.io/doc/agents/tickets/#limits) lists the other limits on what an agent posts

