Brethof AI Reviews

Codex CLI review

A real sandbox, but it reports home by default

Not sponsoredAI Agents · Developer Tools

Tested on Linux · Freemium · Open source

Codex is OpenAI's AI coding agent, and we review the Codex CLI, its open-source terminal version. It edits and runs code in a real Linux sandbox, reviews its own changes, and runs headless in scripts. We tested it on Linux.

The good

  • Open source, Apache 2.0. The whole CLI is open source and public on GitHub, under the Apache licence.
  • A real sandbox. Commands run inside a Linux sandbox by default, with presets from read-only to full access and approval prompts. That's a real sandbox.
  • Reviewer and scripts built in. We like that the reviewer and scripts are built in. The review command checks your changes, and codex review runs it without the interface, for scripts and CI.

The not so good

  • Privacy is opt-out. Codex sends usage data to OpenAI by default and never asks; you have to add a setting to turn it off. With OpenAI, privacy is opt-out.
  • Best with a paid plan. Heads up: with OpenAI models, the CLI needs either a ChatGPT plan, from Plus at twenty dollars a month, or an API key that you pay per use.
  • Other models take setup. Models outside OpenAI take some setup: a provider in the config file that speaks the Responses API. The older settings from many guides now fail, though the errors say how to fix them.

Our verdict

Recommended with caveats

Open source under Apache 2.0, with a real Linux sandbox, a built-in reviewer and a script mode. But it sends usage data to OpenAI without asking, and with OpenAI models it needs a paid ChatGPT plan or a pay-per-use API key.

Reviewed: Codex CLI 0.162.1 · facts checked 2026-10-15

← All reviews

Hear it when it ships

New releases, real benchmarks and the occasional deep-dive. No spam, unsubscribe in one click.

Everything we build

External:   YouTube · GitHub