<div id="readability-page-1" class="page"><div>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.
<h2 id="what-well-build">What we’ll build</h2>
<ul>
<li>Spin up an E2B sandbox from the prebuilt openai-codex template.</li>
<li>Inject your OpenAI API key into the sandbox.</li>
<li>Run a Codex prompt that creates a Hello World file.</li>
<li>Print the output and shut the sandbox down.</li>
</ul>
<h4 id="see-full-example-in-the-e2b-cookbook">See <a href="https://github.com/e2b-dev/e2b-cookbook/tree/main/examples/openai-codex-in-sandbox-python" target="_blank" rel="noopener noreferrer" data-external="true">full example</a> in the E2B Cookbook.</h4>
<h2 id="components-well-use">Components we’ll use</h2>
<ul>
<li>E2B Python SDK – creates and controls the sandbox.</li>
<li>Prebuilt template: openai-codex – Codex CLI is preinstalled.</li>
<li>OpenAI API – used by Codex via OPENAI_API_KEY.</li>
</ul>
<h2 id="prerequisites">Prerequisites</h2>

<h2 id="outline">Outline</h2>
<ol>
<li>Create the project</li>
<li>Add the files (pyproject.toml, main.py)</li>
<li>Set E2B_API_KEY in .env</li>
<li>Install with pip and run</li>
</ol>
<h2 id="1-create-the-project">1) Create the project</h2>
<figure><div><pre tabindex="0"><code>mkdir openai-codex-in-sandbox-python
cd openai-codex-in-sandbox-python
mkdir -p src/openai_codex_in_sandbox_python</code></pre></div></figure>
Create and commit <code>.gitignore</code> before <code>.env</code> so secrets never enter git history.
Target structure:
<figure><div><pre tabindex="0"><code>openai-codex-in-sandbox-python/
├─ src/
│ └─ openai_codex_in_sandbox_python/
│ └─ main.py
├─ pyproject.toml
├─ .gitignore
└─ .env</code></pre></div></figure>
<h2 id="2-files-to-add">2) Files to add</h2>
<h4 id="file-pyprojecttoml">File: <code>pyproject.toml</code></h4>
<figure><div><pre tabindex="0"><code>[project]
name = "openai-codex-in-sandbox-python"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = "&gt;=3.11"
dependencies = [
 "e2b",
 "python-dotenv",
]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"</code></pre></div></figure>
<h4 id="file-srcopenaicodexinsandboxpythonmainpy">File: <code>src/openai_codex_in_sandbox_python/main.py</code></h4>
This script loads <code>.env</code> (so the SDK sees <code>E2B_API_KEY</code>), creates the sandbox from the <code>openai-codex</code> 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:
<figure><div><pre tabindex="0"><code>from dotenv import load_dotenv
from e2b import Sandbox

load_dotenv()</code></pre></div></figure>
Create the sandbox (OpenAI key passed inline) and log its ID:
<figure><div><pre tabindex="0"><code>template_name = 'openai-codex'
sbx = Sandbox(
 template_name,
 envs={
 "OPENAI_API_KEY": "&lt;your api key&gt;",
 },
 timeout=60 * 5, # 5 minutes; match README (remove this line to match the file exactly)
)
print("Sandbox created", sbx.sandbox_id)</code></pre></div></figure>
Run a Codex prompt that creates a file (no per-command timeout):
<figure><div><pre tabindex="0"><code># 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)</code></pre></div></figure>
Clean up:
<figure><div><pre tabindex="0"><code>sbx.kill()</code></pre></div></figure>
<h4 id="file-gitignore">File: .<code>gitignore</code></h4>
<figure><div><pre tabindex="0"><code>.venv/
.env
__pycache__/
*.pyc</code></pre></div></figure>
<h4 id="file-env">File: .<code>env</code></h4>
<figure><div><pre tabindex="0"><code>E2B_API_KEY="YOUR_E2B_API_KEY"</code></pre></div></figure>
<h2 id="3-set-up-and-install">3) Set up and install</h2>
<figure><div><pre tabindex="0"><code>python -m venv .venv
# macOS / Linux
source .venv/bin/activate
# Windows (PowerShell)
# .venv\Scripts\Activate.ps1

pip install -e .</code></pre></div></figure>
<h2 id="4-run-the-example">4) Run the example</h2>
<figure><div><pre tabindex="0"><code>python src/openai_codex_in_sandbox_python/main.py</code></pre></div></figure>
Expected: console output from Codex. The generated index.html lives inside the sandbox filesystem.
<h4 id="see-full-example-in-the-e2b-cookbook-1">See <a href="https://github.com/e2b-dev/e2b-cookbook/tree/main/examples/openai-codex-in-sandbox-python" target="_blank" rel="noopener noreferrer" data-external="true">full example</a> in the E2B Cookbook.</h4></div></div>

E2B Blog & Customer Stories

Vaclav Mlejnsky

Run OpenAI Codex in an E2B Sandbox from Python: an isolated, long-running environment where the agent safely creates files, installs tools, and executes code.

Python Guide: Run OpenAI Codex in an E2B Sandbox

댓글