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

# Agents

> Create, edit, and share AI assistants configured for a specific purpose.

## What it is

An **agent** is an AI assistant configured for a specific purpose. You define the **base model**, the **instructions** (how it should behave), the **knowledge** it consults, prompt suggestions, **advanced parameters**, and **tools**. Then you just chat with it in the chat — and, if you want, **share it** with colleagues.

## How to access

* Sidebar → **Agents**
* To create: **New → Agent** button (at the top) or the creation button on the Agents page.

<Note>
  The Agents page only appears if the **Workspace** module and **Agents** permission are enabled for you.
</Note>

## Agents Gallery

You can view **all agents created by the Symphony team** in the public gallery and **request access** to the ones you need:

<Card title="Open the Agents Gallery" icon="grid-2" href="https://symphony.fcamara.com/agents-gallery">
  Explore the agent catalog by area at **symphony.fcamara.com/agents-gallery** and request access to the desired agents.
</Card>

## Create an agent

On the **Create agent** screen, fill in the following blocks:

<AccordionGroup>
  <Accordion title="1. Agent information" icon="id-card">
    * **Name** *(required)* — how the agent will be displayed.
    * **Base model** *(required)* — the AI model that gives the agent its "intelligence". Without it, you can't save.
    * **Description** — brief summary of what the agent does.
    * **Image/avatar** — agent photo. Accepts **PNG or JPG up to 5 MB**.
    * **ID** *(required on creation)* — unique agent identifier. Cannot repeat an existing ID.
  </Accordion>

  <Accordion title="2. Instructions" icon="list-check">
    **Agent prompt** field: describe the role, goal, tone of voice, response format, and rules the agent must follow. This is what influences its behavior most. There is a **"Learn more"** link with guidance.
  </Accordion>

  <Accordion title="3. Knowledge" icon="book">
    Connect one or more **knowledge bases** (created on the [Knowledge](/en/guia/base-conhecimento) page) so the agent uses those documents as reference. You can also create a new base without leaving the screen.
  </Accordion>

  <Accordion title="4. Prompt suggestions (optional)" icon="lightbulb">
    Add example questions/commands that will appear as suggestions for users of the agent.
  </Accordion>

  <Accordion title="5. Advanced parameters (optional)" icon="sliders">
    Collapsible block to adjust model behavior (e.g., creativity/temperature and other parameters). Also includes:

    * **Tool selection** — which [tools](/en/guia/ferramentas) the agent can use.
    * **Capabilities** *(administrators only)* — such as vision (images), usage, and citations.
    * **JSON preview** — shows the final agent configuration in technical format.
  </Accordion>
</AccordionGroup>

When finished, click **Save**. A message confirms **"Agent created successfully!"** and you return to the list.

## Edit an agent

In the agent list, open an existing agent to **edit**. The same creation blocks are available. Changes are confirmed with **"Agent updated successfully!"**.

## Share an agent

In the agent editor, use **Share**. Define **access control** — which **users** and **groups** can **view/use** ("Can chat") and who can **edit**.

<Warning>
  Before sharing, the required fields (name, base model, and ID) must be filled in.
</Warning>

## Best practices

* Write **clear and objective instructions**: define role, goal, tone, and what the agent **should not** do.
* Connect **relevant knowledge bases** for answers aligned to your context.
* Give a **name and description** that make the agent's purpose clear to anyone using it.
* Use **prompt suggestions** to guide less experienced users.
* Start with few **advanced parameters**; adjust only if necessary.

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Do I need to know programming to create an agent?">
    No. Creating agents is done through a form. Programming is only needed to create **tools** (see [Tools](/en/guia/ferramentas)).
  </Accordion>

  <Accordion title="Why can't I save?">
    Check that you've filled in **Name**, **Base model**, and **ID**. The ID must be unique — if it already exists, choose another.
  </Accordion>

  <Accordion title="Who can see the agent I created?">
    Only you, until you share it. When sharing, you choose users and groups.
  </Accordion>

  <Accordion title="Where do I use the agent after creating it?">
    In the chat, via the **model selector** at the top of the conversation, or by typing `@` in the message.
  </Accordion>
</AccordionGroup>

## Known limitations

* Every agent requires a **base model** — you cannot save without one.
* The avatar image is limited to **5 MB** (PNG/JPG).
* The **Capabilities** field is only editable by **administrators**.
* The list of available base models depends on the **connections/models** configured by the administrator; models hidden by the company don't appear.
* Creating/editing agents depends on the **Agents** permission in the Workspace.