clone/codex
1
0
Fork 0
mirror of https://github.com/openai/codex.git synced 2026-04-24 14:45:27 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
1fe35e4f7a0e7ee7d56b1ff214f71ae33f552348
codex/codex-cli/README.md
Thibault Sottiaux 1fe35e4f7a docs: refresh codex-cli readme
2025-10-20 14:51:01 -07:00

27 KiB
Raw Blame History

OpenAI Codex CLI

Lightweight coding agent that runs in your terminal

npm i -g @openai/codex

Note

This README focuses on the native Rust CLI. For additional deep dives, see the docs/ folder and the root README.

Codex demo GIF using: codex "explain this codebase to me"

Table of contents

Experimental technology disclaimer

Codex CLI is an experimental project under active development. It is not yet stable, may contain bugs, incomplete features, or undergo breaking changes. We're building it in the open with the community and welcome:

  • Bug reports
  • Feature requests
  • Pull requests
  • Good vibes

Help us improve by filing issues or submitting PRs (see the section below for how to contribute)!

Quickstart

Install globally:

npm install -g @openai/codex

Next, set your OpenAI API key as an environment variable:

export OPENAI_API_KEY="your-api-key-here"

Note: This command sets the key only for your current terminal session. You can add the export line to your shell's configuration file (e.g., ~/.zshrc) but we recommend setting for the session. Tip: You can also place your API key into a .env file at the root of your project:

OPENAI_API_KEY=your-api-key-here

The CLI will automatically load variables from .env (via dotenv/config).

Tip

The CLI ships with OpenAI and local OSS providers out of the box. To add additional providers, edit the [model_providers] table in ~/.codex/config.toml. See Configuration guide for examples.

Run interactively:

codex

Or, run with a prompt as input (and optionally in Full Auto mode):

codex "explain this codebase to me"

codex --full-auto "create the fanciest todo-list app"

That's it - Codex will scaffold a file, run it inside a sandbox, install any missing dependencies, and show you the live result. Approve the changes and they'll be committed to your working directory.

Why Codex?

Codex CLI is built for developers who already live in the terminal and want ChatGPT-level reasoning plus the power to actually run code, manipulate files, and iterate - all under version control. In short, it's chat-driven development that understands and executes your repo.

  • Zero setup - bring your OpenAI API key and it just works!
  • Full auto-approval, while safe + secure by running network-disabled and directory-sandboxed
  • Multimodal - pass in screenshots or diagrams to implement features ✨

And it's fully open-source so you can see and contribute to how it develops!

Security model & permissions

Codex lets you decide how much autonomy the agent receives via the --ask-for-approval flag (or the interactive onboarding prompt). The default is on-request.

Mode (--ask-for-approval …) Auto-approves Escalates to you when…
untrusted Built-in "safe" commands that only read files (ls, cat, sed, etc.) The model proposes writing to disk or running any other command.
on-failure All commands, executed inside the configured sandbox with network access disabled and writes limited to the allowed directories. A command fails in the sandbox and the model wants to retry it without sandboxing.
on-request (default) Whatever the model deems safe; it typically asks you before launching riskier commands or writing files. The model decides it wants confirmation, or the sandbox refuses a command and the model asks to retry.
never Everything, with no escalation. Never; failures are returned straight to the model.

Use codex --full-auto as a shorthand for --ask-for-approval on-failure --sandbox workspace-write. For air-gapped or CI environments that provide their own isolation, --dangerously-bypass-approvals-and-sandbox disables both confirmation prompts and sandboxing—double-check before using it.

Platform sandboxing details

The hardening mechanism Codex uses depends on your OS:

  • macOS 12+ - commands are wrapped with Apple Seatbelt (sandbox-exec).

    • Everything is placed in a read-only jail except for a small set of writable roots ($PWD, $TMPDIR, ~/.codex, etc.).
    • Outbound network is fully blocked by default – even if a child process tries to curl somewhere it will fail.

  • Linux - commands run through the bundled codex-linux-sandbox helper. It combines Landlock filesystem rules with a seccomp filter, mirroring the macOS policy: commands start network-disabled and only the working directory (plus a few temp paths) are writable. You still get escape hatches via the --sandbox flag:

    • --sandbox read-only is ideal for review-only sessions.
    • --sandbox danger-full-access turns the sandbox off. Pair it with --ask-for-approval untrusted if you still want Codex to double-check risky commands.

Containers (Docker/Podman) can still be useful when you want completely reproducible toolchains, GPU access, or custom OS packages. In that case launch the CLI inside your container and keep the built-in sandbox on; it will happily sandbox inside the container.

System requirements

Requirement Details
Operating systems macOS 12+, Ubuntu 22.04+/Debian 12+, or Windows 11 via WSL2
Runtime dependencies None for the packaged binaries (install via npm, Homebrew, or releases)
Git (optional, recommended) 2.39+ for built-in PR helpers
RAM 4-GB minimum (8-GB recommended)

Never run sudo npm install -g; fix npm or use another package manager instead.

CLI reference

Command Purpose Example
codex Launch the interactive TUI codex
codex "..." Seed the interactive session with an opening task codex "fix lint errors"
codex exec "..." Run a non-interactive turn in the current repo codex exec "count the total number of TODO comments"
codex exec --json "..." Stream machine-readable events as JSON Lines codex exec --json --full-auto "update CHANGELOG"
codex exec resume --last "..." Resume the most recent non-interactive session codex exec resume --last "ship the follow-up fix"
codex completion <bash|zsh|fish> Print shell completion script for your shell codex completion bash

Helpful flags: --model/-m, --ask-for-approval/-a, --sandbox/-s, --oss, --full-auto, --config/-c key=value, and --web-search.

Memory & project docs

You can give Codex extra instructions and guidance using AGENTS.md files. Codex looks for AGENTS.md files in the following places, and merges them top-down:

  1. ~/.codex/AGENTS.md - personal global guidance
  2. AGENTS.md at repo root - shared project notes
  3. AGENTS.md in the current working directory - sub-folder/feature specifics

Non-interactive / CI mode

Run Codex head-less in pipelines. Example GitHub Action step:

- name: Update changelog via Codex
  run: |
    npm install -g @openai/codex
    export OPENAI_API_KEY="${{ secrets.OPENAI_KEY }}"
    codex exec --json --full-auto "update CHANGELOG for next release" > codex.log

codex exec streams its progress to stderr and writes the final assistant reply to stdout. Use --json when you need structured output, or -o path/to/result.json to capture just the closing message.

Tracing / verbose logging

Set RUST_LOG to control structured logging. The default filter is codex_core=info,codex_tui=info,codex_rmcp_client=info. To turn on verbose logs for troubleshooting:

RUST_LOG=codex_core=debug,codex_tui=debug codex

Logs are written to ~/.codex/logs/codex-tui.log in addition to stderr. You can use standard env_logger syntax (e.g., RUST_LOG=info,reqwest=trace).

Recipes

Below are a few bite-size examples you can copy-paste. Replace the text in quotes with your own task. See the prompting guide for more tips and usage patterns.

✨ What you type What happens
1 codex "Refactor the Dashboard component to React Hooks" Codex rewrites the class component, runs npm test, and shows the diff.
2 codex "Generate SQL migrations for adding a users table" Infers your ORM, creates migration files, and runs them in a sandboxed DB.
3 codex "Write unit tests for utils/date.ts" Generates tests, executes them, and iterates until they pass.
4 codex "Bulk-rename *.jpeg -> *.jpg with git mv" Safely renames files and updates imports/usages.
5 codex "Explain what this regex does: ^(?=.*[A-Z]).{8,}$" Outputs a step-by-step human explanation.
6 codex "Carefully review this repo, and propose 3 high impact well-scoped PRs" Suggests impactful PRs in the current codebase.
7 codex "Look for vulnerabilities and create a security review report" Finds and explains security bugs.

Installation

From npm (Recommended) 
npm install -g @openai/codex
# or
yarn global add @openai/codex
# or
bun install -g @openai/codex
# or
pnpm add -g @openai/codex
Build from source 
# Clone the repository and navigate to the workspace root
git clone https://github.com/openai/codex.git
cd codex

# Ensure you have the latest stable Rust toolchain
rustup default stable

# (Optional) install just for handy automation
cargo install just

# Build the interactive CLI
cargo build -p codex-tui

# Run it directly from source
cargo run -p codex-tui -- --help

Configuration guide

Codex reads configuration from ~/.codex/config.toml (or $CODEX_HOME/config.toml). TOML is the only supported format. Command-line flags (--model, --ask-for-approval, --config key=value, etc.) override whatever is set in the file.

Basic configuration parameters

Key Type Default Description
model string gpt-5-codex (macOS/Linux) / gpt-5 (WSL) Selects the default model.
model_provider string openai Picks an entry from the [model_providers] table.
approval_policy string on-request Matches the CLI --ask-for-approval flag (untrusted, on-failure, on-request, never).
sandbox_mode string workspace-write on trusted repos, otherwise read-only Controls how shell commands are sandboxed (read-only, workspace-write, danger-full-access).
notify array unset Optional notifier command: e.g. notify = ["terminal-notifier", "-message", "Codex done"].
tui_notifications table {"approvals": true, "turns": true} Controls OSC 9 terminal notifications.
history.persistence string save-all save-all, commands-only, or none.
hide_agent_reasoning bool false Suppress reasoning summaries in the UI.

Use codex --config key=value to experiment without editing the file. For example, codex --config approval_policy="untrusted".

Managing model providers

The CLI bundles two providers: openai (Responses API) and oss (local models via Ollama). You can add more by extending the model_providers map. Entries do not replace the defaults; they are merged in.

model = "gpt-4o"
model_provider = "openai-chat"

[model_providers.openai-chat]
name = "OpenAI (Chat Completions)"
base_url = "https://api.openai.com/v1"
wire_api = "chat"
env_key = "OPENAI_API_KEY"

[model_providers.ollama]
name = "Ollama"
base_url = "http://localhost:11434/v1"

Set API keys by exporting the environment variable referenced by each provider (env_key). If you need to override headers or query parameters, add http_headers, env_http_headers, or query_params within the provider block. See docs/config.md for more examples, including Azure and custom retries.

History, profiles, and overrides

  • History is controlled via the [history] table. Example:

    [history]
persistence = "commands-only"
redact_patterns = ["api_key=*"]

  • Use profiles to store alternative defaults:

    [profiles.ops]
model = "gpt-5"
approval_policy = "untrusted"
sandbox_mode = "read-only"

    Launch with codex --profile ops.

  • Override individual keys for a single run: codex --config history.persistence="none".

MCP servers and instructions

Add MCP integrations with [mcp_servers.<id>] blocks (stdio or streamable HTTP). Refer to docs/config.md#mcps for the schema.

For persistent guidance, create AGENTS.md files in ~/.codex, your repo root, or subdirectories. Codex merges them from root to current directory before each turn.

Example config.toml

model = "gpt-5-codex"
model_provider = "openai"
approval_policy = "untrusted"
sandbox_mode = "workspace-write"

[history]
persistence = "save-all"

[model_providers.azure]
name = "Azure"
base_url = "https://YOUR_RESOURCE_NAME.openai.azure.com/openai"
env_key = "AZURE_OPENAI_API_KEY"
wire_api = "responses"
query_params = { api-version = "2025-04-01-preview" }

Restart Codex (or run the next command with --config) after editing the file to pick up changes.

FAQ

OpenAI released a model called Codex in 2021 - is this related?

In 2021, OpenAI released Codex, an AI system designed to generate code from natural language prompts. That original Codex model was deprecated as of March 2023 and is separate from the CLI tool.

Which models are supported?

Any model available via the Responses API. The default is gpt-5-codex (or gpt-5 on Windows/WSL), but pass --model or set model = "gpt-4.1" in config.toml to override.

Why does o3 or o4-mini not work for me?

It's possible that your API account needs to be verified in order to start streaming responses and seeing chain of thought summaries from the API. If you're still running into issues, please let us know!

How do I stop Codex from editing my files?

Run with codex --ask-for-approval untrusted or codex --sandbox read-only to force Codex to ask before making changes. In interactive sessions, you can also deny a specific command or patch by answering n when prompted.

Does it work on Windows?

Not natively. Use Windows Subsystem for Linux (WSL2) and install the Linux build inside your WSL environment. We regularly test on macOS and Linux.

Zero data retention (ZDR) usage

Codex CLI does support OpenAI organizations with Zero Data Retention (ZDR) enabled. If your OpenAI organization has Zero Data Retention enabled and you still encounter errors such as:

OpenAI rejected the request. Error details: Status: 400, Code: unsupported_parameter, Type: invalid_request_error, Message: 400 Previous response cannot be used for this organization due to Zero Data Retention.

You may need to upgrade to a more recent version with: npm i -g @openai/codex@latest

Codex open source fund

We're excited to launch a $1 million initiative supporting open source projects that use Codex CLI and other OpenAI models.

  • Grants are awarded up to $25,000 API credits.
  • Applications are reviewed on a rolling basis.

Interested? Apply here.

Contributing

This project is under active development and we currently prioritize external contributions that address bugs or security issues. If you are proposing a new feature or behavior change, please open an issue first and get confirmation from the team before investing significant effort.

We care deeply about reliability and long-term maintainability, so the bar for merging code is intentionally high. Use this README together with the canonical contributor guide.

Development workflow

  • Create a topic branch from main (for example feat/improve-sandbox).
  • Keep changes focused; unrelated fixes should land as separate PRs.
  • Install Rust 1.80+ and just. Most commands run from the repo root:
    • just fmt formats all Rust code.
    • just fix -p codex-tui runs cargo clippy --fix and cargo fmt for the TUI crate (swap the crate name as appropriate).
    • cargo test -p codex-tui or other crate-specific test commands keep feedback fast.
  • If you touch shared crates (for example codex-core or codex-common), prefer cargo test --all-features after the targeted suite passes.

Debugging

  • Run cargo run -p codex-tui -- to launch the CLI under your debugger of choice. cargo run -p codex-cli --bin codex-linux-sandbox -- --help is helpful when iterating on the sandbox helper.
  • Set RUST_LOG=codex_core=debug,codex_tui=debug to capture verbose logs (see Tracing).
  • Use cargo test -p <crate> -- --nocapture to see println!/tracing output from tests while iterating on new features.

Writing high-impact code changes

  1. Start with an issue. Open a new one or comment on an existing discussion so we can agree on the solution before code is written.
  2. Add or update tests. Every new feature or bug-fix should come with test coverage that fails before your change and passes afterwards. 100% coverage is not required, but aim for meaningful assertions.
  3. Document behaviour. If your change affects user-facing behaviour, update the README, inline help (codex --help), or relevant example projects.
  4. Keep commits atomic. Each commit should compile and the tests should pass. This makes reviews and potential rollbacks easier.

Opening a pull request

  • Fill in the PR template (or include similar information) – What? Why? How?
  • Run all checks locally (cargo test, cargo clippy --tests, cargo fmt -- --check, plus any just fix -p <crate> you relied on). CI failures that could have been caught locally slow down the process.
  • Make sure your branch is up-to-date with main and that you have resolved merge conflicts.
  • Mark the PR as Ready for review only when you believe it is in a mergeable state.

Review process

  1. One maintainer will be assigned as a primary reviewer.
  2. We may ask for changes - please do not take this personally. We value the work, we just also value consistency and long-term maintainability.
  3. When there is consensus that the PR meets the bar, a maintainer will squash-and-merge.

Community values

  • Be kind and inclusive. Treat others with respect; we follow the Contributor Covenant.
  • Assume good intent. Written communication is hard - err on the side of generosity.
  • Teach & learn. If you spot something confusing, open an issue or PR with improvements.

Getting help

If you run into problems setting up the project, would like feedback on an idea, or just want to say hi - please open a Discussion or jump into the relevant issue. We are happy to help.

Together we can make Codex CLI an incredible tool. Happy hacking! 🚀

Contributor license agreement (CLA)

All contributors must accept the CLA. The process is lightweight:

  1. Open your pull request.

  2. Paste the following comment (or reply recheck if you've signed before):

    I have read the CLA Document and I hereby sign the CLA

  3. The CLA-Assistant bot records your signature in the repo and marks the status check as passed.

No special Git commands, email attachments, or commit footers required.

Quick fixes

Scenario Command
Amend last commit git commit --amend -s --no-edit && git push -f

The DCO check blocks merges until every commit in the PR carries the footer (with squash this is just the one).

Releasing codex

To stage npm artifacts for a release, run the helper from the repo root:

./scripts/stage_npm_packages.py \
  --release-version 0.6.0 \
  --package codex

The script assembles native binaries, hydrates the vendor/ tree, and writes tarballs to dist/npm/. Inspect the generated package contents (for example by extracting them or running npm pack --dry-run). When satisfied:

cd dist/npm
npm publish codex-0.6.0.tgz

Add additional --package flags if you need to ship the responses proxy or SDK in the same release. See codex-cli/scripts/README.md for details and troubleshooting tips.

Alternative build options

Nix flake development

Prerequisite: Nix >= 2.4 with flakes enabled (experimental-features = nix-command flakes in ~/.config/nix/nix.conf).

Enter a Nix development shell:

# Use either one of the commands according to which implementation you want to work with
nix develop .#codex-cli # For entering codex-cli specific shell
nix develop .#codex-rs # For entering codex-rs specific shell

This shell includes Node.js, installs dependencies, builds the CLI, and provides a codex command alias.

Build and run the CLI directly:

# Use either one of the commands according to which implementation you want to work with
nix build .#codex-cli # For building codex-cli
nix build .#codex-rs # For building codex-rs
./result/bin/codex --help

Run the CLI via the flake app:

# Use either one of the commands according to which implementation you want to work with
nix run .#codex-cli # For running codex-cli
nix run .#codex-rs # For running codex-rs

Use direnv with flakes

If you have direnv installed, you can use the following .envrc to automatically enter the Nix shell when you cd into the project directory:

cd codex-rs
echo "use flake ../flake.nix#codex-cli" >> .envrc && direnv allow
cd codex-cli
echo "use flake ../flake.nix#codex-rs" >> .envrc && direnv allow

Security & responsible AI

Have you discovered a vulnerability or have concerns about model output? Please e-mail security@openai.com and we will respond promptly.

License

This repository is licensed under the Apache-2.0 License.

Reference in New Issue View Git Blame Copy Permalink