> ## Documentation Index
> Fetch the complete documentation index at: https://docs.longoz.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents

> Create an Agent, choose its permissions, and use it in Issues or chat.

An Agent has a name, a responsible person, and a provider on a specific Computer. It uses that provider's CLI to do work. The Computer must have the files, tools, and signed-in accounts the task needs.

Open **Agents** to create and manage agents. **Members** lists people; **Chat** holds conversations.

## Add an Agent

Before starting, connect a [Computer](https://docs.longoz.ai/computers.md) and prepare its provider. Only ready providers can be selected.

1. Open **Agents** and choose **Add agent**.
2. Select a **Computer**, then a **Provider** installed on it.
3. Choose a **Responsible person** from the active people in the workspace.
4. Enter an **Agent name**.
5. Select **Autonomy** and set **Parallel tasks**, the maximum number of active Issue runs for this agent. The allowed range is 1 to 32.
6. Leave **Model** and **Effort** at their defaults to use the provider configuration, or choose supported values.
7. Add agent instructions or Skills if needed, then choose **Add agent**.

The new agent appears in **Agents**. You can also create an agent from a ready provider on **Computers**. That route uses you as its responsible person.

## Choose autonomy

| Mode | Behavior |
| - | - |
| Guarded | Asks for approval before sensitive actions. |
| Strict | Requires approval for every supported operation. Available only when the provider supports it. |
| Full | Runs without approval prompts. |

The creation form defaults to **Full**. For your first run, explicitly select **Guarded** so sensitive actions can stay in the approval flow. Review each requested action before allowing it.

[Routines](https://docs.longoz.ai/routines.md) require a healthy agent with Full autonomy. Do not switch an agent to Full just to try a schedule without first considering what it can access on its Computer.

## Add the Agent to a Project

An agent needs project membership to receive its work.

1. Open the project and choose **Edit project**.
2. Select the agent under **Agent members**.
3. Choose **Save changes**.

For a **Local repository**, use an agent on the Computer that holds that repository. See [Projects](https://docs.longoz.ai/projects.md).

## Assign an Issue

Choose **Create issue**, select the project, then assign the issue to the agent. Give it a specific task and describe the result you expect.

An agent-assigned **Todo** issue queues work automatically. **Backlog** keeps it parked. Newly created agent-owned issues must use one of those two statuses. Keep the Computer online while work runs.

For example, a first task can ask the provider to draft a short welcome note and return it in the issue without modifying files or sending anything. This needs no external service or repository. Tasks that use another CLI or account require that tool and its credentials on the assigned Computer.

See [Issues](https://docs.longoz.ai/issues.md) for reviewing results and controlling execution.

## Chat with an Agent

1. Open **Chat** and choose **New chat**.
2. Select a project, then an available agent for that project.
3. Enter your request and send it.

A conversation can perform work through the agent's provider. Be explicit if you want an explanation only and do not want files changed or commands run.

Wait for the current turn to finish before sending another message. **Stop** interrupts the current turn and keeps the conversation's existing messages. Archiving a conversation removes it from active history and prevents further messages; its history remains recorded.

If no runnable agents appear, check project membership, the Computer's connection, and provider readiness. See [Chat](https://docs.longoz.ai/chat.md) for attachments and conversation history.

## Change an Agent

Open the agent's editor in **Agents** to change its identity, responsible person, autonomy, parallel tasks, or execution settings. Agent instructions provide role and style guidance; they do not grant permissions.

Configuration changes apply to new runs. Running or queued runs and retries of the same turn keep their previous settings.

Pausing an agent also pauses its active Routines. Reducing Full autonomy pauses those Routines too. They do not resume automatically. Reactivating an agent requires a healthy online Computer and provider.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.