Guided Codex login
Reuses your existing Codex session, or opens the official browser login only when authentication is needed.
Free with open source code · MIT licensed
Claude and Codex models.
Claude and Codex models.
One launcher.
Run native Claude models or Codex GPT models through separate provider processes. Start both in separate terminals when you want them working at the same time.
macOS
Linux
Windows
WSL
The whole point
Free, open download.
One command installs Claudex from its verified source release and sets up everything it needs.
MIT license
Open source
macOS · Linux · Windows
curl -fsSL --proto '=https' --tlsv1.2 https://claudex.work/install.sh | bash
Verified source install · macOS, Linux, and WSL
irm https://claudex.work/install.ps1 | iex
Run in PowerShell · verified Windows source install
The installer adds Claude Code and the Codex CLI when needed, opens the official Codex browser login when authentication is required, and verifies release downloads with SHA-256. Claudex then keeps itself up to date. See the README for details.
See it in action
Built for real Claude Code terminal sessions
Choose native Claude or managed GPT
Start native models with claudex --fable, --opus,
--sonnet, or --haiku. Use claudex --claude-model MODEL
for any alias or full identifier. The managed route provides Sol, Terra, Luna, and Solplan.
Fableplan plans with native Fable, then starts managed Terra with only the returned plan text.
Know your usage before you hit it
The status line never blocks on network I/O. It reads a cached, sanitized
summary and refreshes it in the background. Run /usage-limit any time
for the full breakdown, straight from your Codex account.
Verified, isolated, and honest about it
claudex --doctor reports bridge health, proxy version, and
authentication state at a glance and never prints a token while doing it.
Everything it adds
A bridge, not a fork
Native Claude models
Fable, Opus, Sonnet, Haiku, and full model identifiers pass to your installed Claude Code and provider.
Managed Codex models
Sol, Terra, Luna, and Solplan map onto Codex's GPT 5.6 models for the work each is best at.
Two planning workflows
Solplan uses Sol then Terra. Fableplan uses a native read only Fable planner, then isolated Terra implementation.
Concurrent provider sessions
Run Claude and GPT at the same time in separate terminals without sharing provider credentials or session state.
Stable context accounting
Tracks context honestly and compacts automatically near 280k tokens with no false zeroes.
Usage limits in the status line
Five hour and seven day Codex windows, always visible, with a full report on demand.
Works on every major platform
macOS, Linux, Windows, and WSL. Bash and PowerShell stay in behavioral parity.
Setup with one command
Installs missing Claude Code and Codex CLI dependencies, then checks the complete local bridge.
Automatic updates
Checks for new releases in the background and applies SHA-256-verified updates without interrupting your work.
Under the hood
How Claudex works
A launcher, not a fork. Each provider gets its own process and credentials. Nothing about the signed Claude Code binary ever changes.
Managed GPT route
You
run
claudex
Claude Code
terminal UI
CLIProxyAPI
127.0.0.1:8318
Codex
your account
Native Claude route
You
run
claudex --fable
Claude Code
clean process
Your provider
caller configuration
Claude
your model access
Security posture
Safe on your own machine
Never patches or forks the signed Claude Code executable.
The local bridge binds to 127.0.0.1 only, guarded by a generated key.
Every downloaded binary is pinned by version and verified by SHA-256.
Credentials live under ~/.config/claudex, never in arguments, logs, or Git.
Questions
Frequently asked
What is Claudex?
Claudex is an open source launcher for native Claude models and Codex GPT models. Claude and GPT routes use separate provider processes, and Claudex leaves both Claude Code and the Codex CLI as their official, unmodified installs.
Do I need an OpenAI API key?
Claudex does not require an OpenAI API key. Claudex reuses the Codex login already on your computer, and opens Codex's official browser login only when authentication is needed. You never create provider keys or configure model endpoints by hand.
Does Claudex work with a ChatGPT Plus or Pro plan?
Claudex works with whatever your current Codex account provides, and the startup banner
reports the tier it detects: ChatGPT Free, Go, Plus, Pro, or the applicable workspace
tier. Model availability follows the account, so run claudex --doctor to see
exactly which models your account advertises.
Which models does Claudex provide?
Claudex supports native Fable, Opus, Sonnet, Haiku, and full Claude model identifiers accepted by your installed Claude Code and provider. Its managed Codex route provides Sol, Terra, Luna, and Solplan. Availability still follows your accounts and upstream services.
Can I run Claude and GPT models at the same time?
Yes. Start a native Claude route in one terminal and a managed GPT route in another. Claudex keeps them in separate processes with separate provider credentials and session state instead of changing providers inside one Claude Code process.
What is Fableplan?
Fableplan runs native Fable as a read only planner, then starts managed Terra to implement the returned plan. The two providers stay in separate processes, and only the returned plan text crosses through a private temporary file.
Does Claudex modify or fork Claude Code?
Claudex never patches or forks the signed Claude Code executable. Claudex is a launcher: it
starts an isolated Claude Code profile, keeps its generated configuration under
~/.config/claudex, and leaves your normal Claude Code settings untouched.
Is Claudex safe to install?
Claudex runs entirely on your own machine. Its local bridge binds to 127.0.0.1 only and is
guarded by a generated key, every downloaded binary is pinned by version and verified by
SHA-256, and credentials stay under ~/.config/claudex, never in arguments,
logs, or Git. Claudex is MIT licensed and its complete source is on GitHub.
Which platforms does Claudex support?
Claudex supports macOS 13+ on Apple silicon or Intel, Ubuntu 20.04+, Debian 10+ and compatible Linux on x64 or ARM64, Windows 10 1809+, Windows 11 and Windows Server 2019+ on x64 or ARM64, and WSL 1 or WSL 2. Claude Code's own platform limitations still apply.
Can I install Claudex with Homebrew or Scoop?
Claudex ships on both. On macOS or Linux, run brew install BeamoINT/tap/claudex.
On Windows, run scoop bucket add beamoint https://github.com/BeamoINT/scoop-bucket
followed by scoop install beamoint/claudex. WinGet is not currently an installable
channel.
How do I update or uninstall Claudex?
Claudex checks its stable release channel once a day without blocking startup and applies
SHA-256-verified updates by default; claudex self-update --status,
--check, and --apply drive it directly. To uninstall Claudex, close
any active sessions, then remove the claudex launcher and its private
configuration directory ~/.config/claudex.
Is Claudex affiliated with OpenAI or Anthropic?
Claudex is an independent community project. Claudex is not affiliated with, endorsed by, or supported by OpenAI or Anthropic, and you remain responsible for the terms and usage limits of the accounts it connects to.
Ready to run Codex in Claude Code?
Free, open source, and installed entirely on your own machine.