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

# Agents

> What agents are in UtopikAI and when to use them

**Agents** are for work that runs **on its own**, on a **schedule**, or inside a **workflow**. Agents also have a chat UI, but they are built for autonomous or automated runs — not only live back-and-forth conversation.

Each agent is its own entity with its own instructions, models, knowledge, skills, tools, MCP servers, prompts, and optional tasks. Changing one agent does not affect assistants or other agents.

## Custom agents

Open **Custom Agents** from the product to create or edit agents. Built-in agents may also be available depending on your tenant.

Agents can be attached to **workflow** steps and assigned to **tasks** so work runs without someone staying in chat.

## What you can configure

| Area               | What it does                                                            |
| ------------------ | ----------------------------------------------------------------------- |
| **Instructions**   | Main prompt that defines role, tone, and how the agent should act       |
| **Models**         | One or more LLMs, including a default model for runs                    |
| **Knowledge**      | Document sets and search options the agent can reference                |
| **Skills**         | Reusable capabilities attached to this agent only                       |
| **Tools**          | Connected apps for fetching data, taking actions, or generating content |
| **MCP servers**    | External systems connected through Model Context Protocol — agents only |
| **Prompt Library** | Reusable `/` prompts for chat or runs                                   |
| **Access**         | Who in the Workspace can use the agent                                  |
| **Tasks**          | Scheduled or on-demand runs without a live chat session                 |

<Note>
  **MCP servers** attach to **agents**, not assistants. Each agent has its own MCP server list.
</Note>

## When to use an agent

Create an **agent** when work should run unattended, repeat on a schedule, or chain with other steps in a workflow — for example generating reports, processing files, or calling external systems through MCP.

If people will primarily use interactive Q\&A in chat, create an **assistant** instead.

<Card title="Creating Agents" icon="wand-magic-sparkles" href="/guides/agents">
  Step-by-step guide to the eight-step agent setup wizard.
</Card>

## Related documentation

* [Creating Agents](/guides/agents) — Full setup wizard
* [How agents differ from assistants](/guides/assistants_vs_agents) — When to use each
* [Assistants](/guides/assistants_overview) — Interactive chat
* [MCP Servers](/mcp/overview)
* [Workflows Overview](/workflows/overview)
* [Tasks Overview](/tasks/overview)
* [Skills Overview](/skills/overview)
* [Tools Overview](/tools/overview)
* [Introduction](/introduction)