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

# Anaconda Agent Studio

export const DefinitionDescription = ({children}) => <dd className="definition-description">{children}</dd>;

export const DefinitionTerm = ({children}) => <dt className="definition-term">{children}</dt>;

export const DefinitionList = ({children}) => <dl className="definition-list">{children}</dl>;

<Note>
  Anaconda Agent Studio is a beta feature of Anaconda Desktop. Features and functionality might change during the beta period.
</Note>

Anaconda Agent Studio is a desktop application for creating, configuring, and running AI agents locally on your machine. You can build agents through a visual interface or by using the built-in Anaconda Assistant, then connect them to external services, your own tools, and the AI models of your choice.

## Installing Agent Studio

<Note>
  <Icon icon="robot" iconType="solid" /> **Agent Studio** is displayed in the left-hand navigation by default. Disable it on the [Settings](/tools/anaconda-desktop/settings#enabling-beta-features) page. If you belong to an organization, your administrator may restrict access to beta features.
</Note>

1. In the left-hand navigation, under <Icon icon="flask" iconType="solid" /> **BETA**, select <Icon icon="robot" iconType="solid" /> **Agent Studio**.
2. On the Agent Studio page, click <Icon icon="download" iconType="solid" /> **Download**.
3. Accept the terms of service and privacy policy to enable the download.
4. Click <Icon icon="download" iconType="solid" /> **Download** again to begin the installation.

   <Note>
     Download progress displays in the upper-right corner. A notification appears when the download completes.
   </Note>

## Launching Agent Studio

1. From the <Icon icon="robot" iconType="solid" /> **Agent Studio** page, click <Icon icon="arrow-up-right-from-square" iconType="solid" /> **Launch**.
2. Click **Sign in** to authenticate. A browser window opens to complete sign-in with your Anaconda account credentials.

## Capabilities

<DefinitionList>
  <DefinitionTerm>Agent creation</DefinitionTerm>
  <DefinitionDescription>Create agents from templates or from scratch. Each agent can be configured with a system prompt, Python tools, plugins, MCP servers, skills, and reference documents.</DefinitionDescription>

  <DefinitionTerm>AI model providers</DefinitionTerm>
  <DefinitionDescription>Connect to an AI provider and select from its available models. Agent Studio includes Anaconda's hosted models by default. You can also add your own API keys from providers like OpenAI, Anthropic, Google, or Groq, or connect to a model running locally in Anaconda Desktop.</DefinitionDescription>

  <DefinitionTerm>Tools, plugins, and MCP servers</DefinitionTerm>
  <DefinitionDescription>Write custom Python tools for your agents, add plugins that bundle integrations for services like Atlassian and Miro, or connect to any MCP server for access to external APIs and services. Every agent is also exposed as an MCP server, so agents can delegate tasks to each other.</DefinitionDescription>

  <DefinitionTerm>Compare chat</DefinitionTerm>
  <DefinitionDescription>Send the same prompt to up to four agents or models simultaneously and compare their responses side by side.</DefinitionDescription>

  <DefinitionTerm>Sandbox isolation</DefinitionTerm>
  <DefinitionDescription>Run agents inside Docker containers to isolate them from your local system. Configure network access, resource limits, and folder mounts per agent.</DefinitionDescription>

  <DefinitionTerm>Safety guardrails</DefinitionTerm>
  <DefinitionDescription>The agent runtime automatically screens messages for PII, prompt injection, and secrets. When a safety filter triggers, the response is blocked and Agent Studio displays a warning in the chat.</DefinitionDescription>

  <DefinitionTerm>File-based configuration</DefinitionTerm>
  <DefinitionDescription>Agent configurations are stored as local files that you can version control with Git or share with teammates.</DefinitionDescription>
</DefinitionList>