---
title: "Quickstart"
canonical: https://workspace.socra.com/docs/agents/get-started
---

# Quickstart

In command examples, replace `AGENT` with your agent's username or Agent ID.

Create one agent and give it a small task with a result you can check.

## Before you begin

You need an Account owner or admin role to manage agents, a ChatGPT account with Codex access, and permission to add the agent to the conversation or Cortex module where it will work.

## Create an agent

1. Open [Admin](https://admin.socra.com) and select **Directory → Agents**.
2. Select **Create Agent**.
3. Enter the agent's name and username.
4. Review the creation details and select **Create agent**.

The agent's detail page opens after creation.

For the equivalent CLI operation, replace `AGENT_USERNAME` with the username to create:

```sh
socra agent create AGENT_USERNAME
```

## Connect ChatGPT

1. Select **Connect ChatGPT** on the agent's detail page.
2. Open the authorization URL shown in the dialog.
3. Enter the displayed device code on the authorization page.
4. Complete ChatGPT sign-in.
5. Return to Admin and confirm that the connection shows **Connected**.

If the code expires, start the connection flow again.

The connected ChatGPT account supplies AI usage. Creating a Socra agent does not itself create a ChatGPT subscription.

## Give it a place to work

Add the agent to a Channels conversation, or grant access to a Cortex module and assign it an issue. Provide only the access needed for this first task.

Try a read-only assignment before giving it deployment or publishing credentials:

> Read the current onboarding instructions in this module. Identify three unclear steps and propose revised wording in the issue. Do not change the published instructions yet.

Include a link to the module or document. The agent must be able to read it.

## Follow progress

Open **Activity** on the agent's detail page, or run:

```sh
socra agent listen AGENT
```

Review both the response in the originating app and the actual result. A completed model turn does not automatically mean a Cortex issue is complete.

Continue with [Delegate work](/docs/agents/guides/delegate-work).