Codex
OpenAI's open-source terminal coding agent with sandboxed execution and a unified CLI, IDE, and cloud platform
About
Codex is OpenAI's AI coding agent for software engineering tasks — writing code, fixing bugs, and executing multi-step development workflows. The open-source Codex CLI (Apache-2.0, written in Rust) runs locally in your terminal: it reads and modifies files, executes commands in a sandboxed environment, and performs tasks through an interactive TUI or headless 'codex exec' mode for scripts and CI.
The CLI is one surface of the broader Codex platform, alongside IDE extensions, a desktop app, and Codex Web — all share the Codex name, but only the CLI is fully open source. Sign-in works via a ChatGPT account (included across ChatGPT's plans per independent reviews) or an OpenAI API key for usage-based billing in automation. We did not hands-on verify the current default model lineup — launch-era docs cite GPT-4.1 and o3/o4-mini, while newer docs show different model names — treat model claims as current-docs-verified only.
What it can do
- • Reads and edits files, runs commands and tests in a sandboxed execution environment with approval-gated tool use
- • Interactive TUI plus headless 'codex exec' mode for scripts, pipelines, and CI
- • AGENTS.md project instructions; per-session control of model, reasoning effort, and permissions (/model, /permissions, /status)
- • Subagents for delegating complex investigations; Skills/Plugins for reusable instructions and team integrations
- • MCP server support for connecting external tools and data sources; lifecycle hooks for customization
- • 'codex review' examines uncommitted changes, commits, or branches without modifying the working tree
- • Codex cloud: submit tasks from the CLI to a cloud environment that keeps running after you close your laptop; session sync across devices via ChatGPT relay
Setup
- Install on macOS/Linux: 'curl -fsSL https://chatgpt.com/codex/install.sh | sh'. Windows: run 'powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"'. Alternatives: 'npm install -g @openai/codex' or 'brew install --cask codex'.
- Verify: 'codex --version'.
- In a project directory, run 'codex'. On first launch choose 'Sign in with ChatGPT' (browser flow) — included in ChatGPT plans; or use 'codex login --with-api-key' with OPENAI_API_KEY set for usage-based billing.
- Run '/init' to generate an AGENTS.md instructions file for the project; use '/status', '/permissions', and '/model' to control the session.
- For automation: 'codex exec "..."' in scripts and CI; 'codex review' for read-only review of uncommitted changes or branches; 'codex cloud' submits tasks to the cloud environment.
Prerequisites
macOS, Linux, or Windows; a ChatGPT account (Free plan and up per independent reviews) or an OpenAI API key; an internet connection.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 128,827 GitHub stars (log-scaled vs catalog peers); blended with ~97,857,586 monthly package downloads; last push 0d ago; 100 releases/12mo; 22207 open issues. Manual review: ~22k open issues; external PRs not accepted (repo policy); credentials in ~/.codex/auth.json. License: Apache-2.0.
★ 128,826 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.6/10
- Maintenance · recency, releases, issue load
- 8.2/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · 128,827 GitHub stars (log-scaled vs catalog peers)
- · blended with ~97,857,586 monthly package downloads
- · last push 0d ago; 100 releases/12mo; 22207 open issues
- · manual review: ~22k open issues; external PRs not accepted (repo policy); credentials in ~/.codex/auth.json
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.