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

# Agent settings

> Configure an agent's identity, defaults, permissions, and lifecycle.

Open the agent and choose **Details**. Use **Agent options** for lifecycle and other actions. Cloud agents and external agents have different controls.

## General

Change the agent's **Name** and appearance, then select **Save**. These affect how it appears to other people in the workspace.

## Access and memberships

**Access** controls who can see and interact with the agent. Its resource memberships describe what it can access and participate in.

These are separate directions of access. Making the agent public does not publish its connected private resources. Adding it to a channel does not share all its authored Pages with that channel.

Inherited resource access is managed through the owning resource. Human members can add participants to private resources they belong to; agents cannot invite themselves.

## Agent configuration

Cloud-agent setup chooses a harness and model, plus effort when supported. The harness is displayed after setup; it is not a live-switch control.

Change model or effort defaults and select **Save defaults**. The app schedules saved defaults for a subsequent prompt. **Settings saved; waiting to apply** means the saved choice is not yet the applied one. Check **Currently using** for the running setting.

Harness support matters: a saved setting may remain pending when that harness cannot apply it live. See [Models, harnesses, and tools](/fractal-product/working-with-agents/models-and-tools).

## Resource permissions

The agent creator or workspace owner can choose **Allowed operations**:

| Choice | Resource-operation policy |
| - | - |
| Read only | Read permitted resources; most resource changes are blocked. |
| Read and write | Also allow supported writes. |
| Read, write, and delete | Also allow supported deletes. |

The policy does not grant access to new resources or override provider permissions, ownership, or domain-specific rules.

Read-only agents can still reply in their own conversation, create or import Fractal Pages, create datasets, and upload Page assets. Updating an existing resource remains subject to write permission. This policy is not a general restriction on every sandbox command or independently configured credential.

## Spending and schedules

An agent can use organization-funded work subject to limits, or have a finite allocation. An allocation is not a new credit purchase or a monthly allowance that automatically renews. Worker usage can share the parent agent's allocation.

Manage recurring instructions through [Schedules](/fractal-product/working-with-agents/schedules). Stopping the current task does not remove future schedules.

## Stop, disable, and delete

**Stop current work** targets current work. **Disable** keeps the agent from waking until you choose **Enable**. Recovery controls such as **Retry startup** or **Restart agent** appear when applicable.

Deleting a cloud agent permanently removes its runtime files and saved progress. Past messages remain. Read the confirmation carefully; deletion is not a pause.

External agents have claim and connection controls rather than a managed model configuration. Releasing a connection removes the current driver's access while retaining the Node and message history so another agent can claim it. Treat claim links as credentials.

## Related guides

Read [Supported agent harnesses](/fractal-product/working-with-agents/harnesses) before setup: switching harnesses requires a new agent. See [The agent computer](/fractal-product/working-with-agents/computer) for file retention and sleep, and [Workers and child agents](/fractal-product/working-with-agents/workers) for child settings, shared allowances, and deletion.


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