Python Guide: Run OpenAI Codex in an E2B Sandbox

OpenAI Codex turns natural-language instructions into code edits and files. Run it inside an E2B Sandbox and you get an isolated, long-running environment where the agent can safely create files, install tools, and execute without interactive prompts.

What we’ll build

• Spin up an E2B sandbox from the prebuilt openai-codex template.
  
• Inject your OpenAI API key into the sandbox.
  
• Run a Codex prompt that creates a Hello World file.
  
• Print the output and shut the sandbox down.

See full example in the E2B Cookbook.

Components we’ll use

• E2B Python SDK – creates and controls the sandbox.
  
• Prebuilt template: openai-codex – Codex CLI is preinstalled.
  
• OpenAI API – used by Codex via OPENAI_API_KEY.
  

Prerequisites

• Python 3.11+‍
• E2B account + E2B_API_KEY
  

Outline

1. Create the project
2. Add the files (pyproject.toml, main.py)
3. Set E2B_API_KEY in .env
4. Install with pip and run

1) Create the project

mkdir openai-codex-in-sandbox-python
cd openai-codex-in-sandbox-python
mkdir -p src/openai_codex_in_sandbox_python

Create and commit .gitignore before .env so secrets never enter git history.

Target structure:

openai-codex-in-sandbox-python/
├─ src/
│  └─ openai_codex_in_sandbox_python/
│     └─ main.py
├─ pyproject.toml
├─ .gitignore
└─ .env

2) Files to add

File: pyproject.toml

[project]
name = "openai-codex-in-sandbox-python"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.11"
dependencies = [
    "e2b",
    "python-dotenv",
]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

File: src/openai_codex_in_sandbox_python/main.py

This script loads .env (so the SDK sees E2B_API_KEY), creates the sandbox from the openai-codex template while injecting your OpenAI key, runs a Codex exec prompt that generates index.html, prints the result, and then shuts the sandbox down.

Load configuration and SDK:

from dotenv import load_dotenv
from e2b import Sandbox

load_dotenv()

Create the sandbox (OpenAI key passed inline) and log its ID:

template_name = 'openai-codex'
sbx = Sandbox(
    template_name,
    envs={
        "OPENAI_API_KEY": "<your api key>",
    },
    timeout=60 * 5,  # 5 minutes; match README (remove this line to match the file exactly)
)
print("Sandbox created", sbx.sandbox_id)

Run a Codex prompt that creates a file (no per-command timeout):

# Print help for Codex
# result = sbx.commands.run('codex --help', request_timeout=0, timeout=0)
# print(result.stdout)

# Run a prompt with Codex
result = sbx.commands.run(
    "codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox 'Create a hello world index.html'",
    timeout=0,
)
print(result.stdout)

Clean up:

sbx.kill()

File: .gitignore

.venv/
.env
__pycache__/
*.pyc

File: .env

E2B_API_KEY="YOUR_E2B_API_KEY"

3) Set up and install

python -m venv .venv
# macOS / Linux
source .venv/bin/activate
# Windows (PowerShell)
# .venv\Scripts\Activate.ps1

pip install -e .

4) Run the example

python src/openai_codex_in_sandbox_python/main.py

Expected: console output from Codex. The generated index.html lives inside the sandbox filesystem.

See full example in the E2B Cookbook.