---
title: Codex
description: Run OpenAI's Codex in Glade with your ChatGPT or API account. What it needs, how to sign in, and what works.
sidebar:
  icon: "/agents/codex.svg"
---

Glade runs your installed Codex CLI with the account it's signed in to.

| Requirement     | Value                                                     |
| --------------- | --------------------------------------------------------- |
| Minimum version | 0.158.0                                                   |
| Install         | [Codex CLI](https://developers.openai.com/codex/cli)      |
| Sign in         | `codex login`, or **Sign in** in **Settings → Providers** |

## Set up

1. **Install Codex**

    Follow [OpenAI's guide](https://developers.openai.com/codex/cli), then check that `codex` runs
    in a fresh terminal.

2. **Sign in**

    Run `codex login`, or use **Sign in** next to Codex in **Settings → Providers**.

3. **Pick a model**

    Choose a Codex model in the model picker. Reasoning effort and speed come from Codex itself.

## What works

- **Steering:** a message sent mid-turn can redirect the run.
- **Native forks** from any finished reply.
- **`/review`** runs Codex's built-in code review.
- **Approve for me** uses Codex's own reviewer to approve routine actions.
- **Questions while it works:** Codex can ask without stopping; the questions wait above the composer.
- **Usage** shows the five-hour and weekly windows, plus banked resets you can spend.
- **Continue after a usage limit:** Glade picks the chat back up when your limit resets.
- **Skills** are invoked with `$`, like `$deploy`.

## Good to know

- Codex reads its own [configuration](https://github.com/openai/codex/blob/main/docs/config.md).
- Archiving a chat archives it in Codex too, and deleting one deletes its Codex session.

## Under the hood

Glade talks to Codex through `codex app-server`, the same protocol Codex's own clients use. Codex reads
the shared skills folder itself, and Glade's tools and your [MCP servers](/features/mcp-servers) are
added to each thread's MCP configuration.