# Agents

### What is an agent?

Agents are Sculptor’s core unit of work.

An agent is a self-contained worker that:

* Runs in its **own Docker container**
* Uses your configured **Anthropic API key** and model
* Has its own **Git repository and branch**

  ![Agent overview showing container and branch](https://3805321663-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FjFN3RamrZpFlLcWLQKea%2Fuploads%2Fgit-blob-215f3122cd23ca63ea9f6abc2cb073adc132ec1d%2Fagent-overview.png?alt=media)

You describe a task, and the agent reads files, edits code, runs commands, and commits changes on that branch.

> **Mental model:** Treat each agent like a disposable feature branch: do the work, review it, merge what you like, then archive or delete it.

***

### Where agents appear

When you start a new task, Sculptor:

1. Creates an agent
2. Shows it in the **left sidebar** with an auto-generated name based on your prompt
3. Opens a view for that agent’s conversation, changes, and logs

Each task you start creates exactly one agent.

***

### Agents, Git, and containers

Every agent runs in a dedicated container that includes:

* A copy of your project
* Its own file system
* A Git branch like:

  ```bash
  sculptor/<auto-generated-name>
  ```

  ![Git branch created by agent](https://3805321663-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FjFN3RamrZpFlLcWLQKea%2Fuploads%2Fgit-blob-93abd093ebe957229b284696bf0b4fd93be67e96%2Fagent-git-branch.png?alt=media)

All edits and commits happen **inside this container**, not in your local repo, until you choose to pull them back.

This means you can:

* Let agents do large refactors safely
* Inspect the results in Sculptor or via Pairing Mode
* Decide what to merge and what to discard

***

### What agents can do

Inside their container, agents can:

* Read and understand files across your repo
* Generate or modify code
* Run terminal commands (tests, builds, linters, etc.)
* Create and update Git commits
* Prepare changes for merge into your local branch

  ![Agent making code changes](https://3805321663-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FjFN3RamrZpFlLcWLQKea%2Fuploads%2Fgit-blob-d072496accabf8ef774167a95c457b71d5c6a858%2Fagent-code-changes.png?alt=media)

You can continue prompting in the same agent to refine or extend its work; it keeps the full context of the current task.

***

### Models used by agents

Agents use Anthropic models under the hood. When configuring Sculptor, you:

1. Provide your **Anthropic API key**

![Anthropic API key configuration](https://3805321663-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FjFN3RamrZpFlLcWLQKea%2Fuploads%2Fgit-blob-924475f63060767c9aac3a2142826f08865cce9b%2Fagent-api-key-config.png?alt=media)

You can:

* Change the default model for new agents
* Switch an individual agent to a different model if a task needs more speed or more reasoning

> **Note:** Model availability\
> The exact list of models comes from Anthropic and may change over time. Make sure you're on the latest Sculptor version to see newly supported models. See the [Models](https://docs.imbue.com/core-concepts/core-concepts/models) page for more details.

***

### Managing agents

From the main UI you can:

* Open an agent from the sidebar to inspect its log and changes
* Keep prompting in the same agent to iterate on the same branch
* Delete an agent from its settings menu when you're done

  ![Agent management in sidebar](https://3805321663-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FjFN3RamrZpFlLcWLQKea%2Fuploads%2Fgit-blob-c149dbd6b38693ac60758e6d8094116ff850653a%2Fagent-management.png?alt=media)

> **Important:** Deleting an agent removes its container and branch inside Sculptor. Any changes you already pulled into your local repo remain, but unmerged work in that agent is lost.
