Guide · Open source

How to use codex-chatgpt-web: install, models and setup

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

codex-chatgpt-web is a launcher app that lets Codex, a coding assistant, use the ChatGPT Web models available on your own ChatGPT account, including Pro where your account exposes it. This guide covers installing the launcher, choosing a model, and setting up the optional Full harness mode that gives ChatGPT access to your current task's files, terminal and tools. By the end you will know what the tool does, what it needs, and what risks to weigh before using it.

codex-chatgpt-webmiuuyy/codex-chatgpt-web · ★ 15.4k · MIT Install options & downloads

Quick start

1Download the launcher. Get the installer for your system from the release links, or run the macOS/Linux or Windows PowerShell install command.
2Sign in to ChatGPT. Open the launcher, sign in to ChatGPT in its embedded browser, and run the browser smoke test.
3Install models. Install the models from the launcher, then restart Codex once.
4Choose a (Web) model. In automatic mode, pick a model ending in (Web) in Codex's model picker and set Effort if offered.
5Set up Full harness (optional). Open MCP, create the Tunnel and API key, press Connect harness, then create the Codex Native2 connector in ChatGPT.
6Verify the runtime. Run Verify runtime and, if something fails, run Settings then Run doctor.

Check whether this tool fits your situation

Messages run through your signed-in ChatGPT Web session, so they count against that account's ChatGPT limits rather than Codex quota. The README calls this unofficial browser automation, not an OpenAI API, and says UI changes can break it.

It is a good fit if you already pay for ChatGPT and accept an unofficial integration on a personal machine you trust. It is a poor fit if you need a stable, supported API or share your computer.

Release targets: macOS 13+ (arm64/x64), Windows x64, and Linux x64.
Builds are not yet platform-signed, so Gatekeeper or SmartScreen may warn.
Task prompts are processed by OpenAI and conversations remain in your ChatGPT history.

Install the launcher

The simplest route is to download the installer for your system from the project's release links: a Windows x64 .exe, a macOS Apple silicon .dmg (Intel .dmg also listed), or a Linux x64 .AppImage. The app includes its own browser and runtime, so the README says no separate Chrome, Node or Bun install is needed.

The README also provides terminal installers. Quit the launcher before updating. The installers detect platform and architecture, verify published checksums, and keep your ChatGPT profile and launcher settings.

curl -fsSL https://github.com/miuuyy/codex-chatgpt-web/releases/latest/download/install-launcher.sh | sh

Open the launcher and sign in to ChatGPT in the embedded browser, then run the browser smoke test. Next, install the models and restart Codex once. In automatic mode, choose a model ending in (Web).

Which models appear depends on your account. The README lists Luna / Think for Free and Go accounts, and GPT-6 Sol and GPT-5.6 Sol with Instant to High effort for paid accounts, plus Extra High and Pro when available. Pro versions have separate entries, and Sol reasoning is selected through Effort.

Existing GPT-5.6 chats keep GPT-5.6.
Availability depends on your account and workspace.

Understand the three modes

The launcher offers three modes. Browser-only sends messages automatically but gives no local Codex tools. Full harness (With Automation) sends automatically and gives tools through MCP. Zero Risk never reads or operates the ChatGPT page: you select the model and the Codex Zero Risk connector yourself, paste and send the prepared prompt, then confirm Sent in the launcher.

Browser-only: automatic sending, no local tools.
Full harness: automatic sending, local Codex tools through MCP.
Zero Risk: manual paste and send, local tools through a separate MCP connector.

Set up Full harness mode

Full mode connects ChatGPT's tool calls back to your current Codex task through the official OpenAI tunnel-client. The README says the tunnel is outbound, so it does not expose a public IP, open an inbound port, or need router forwarding. The launcher uses your system proxy for the tunnel.

Open MCP in the launcher and follow its guided setup. Create the Tunnel and a regular API key, then press Connect harness. In ChatGPT, enable Developer Mode and create a Tunnel connector named exactly Codex Native2 with Authentication set to None and Allow all actions. Finally run Verify runtime to confirm the connector is attached and available.

Write or modify actions also need your ChatGPT workspace administrator policy to allow them. Unexpected approval prompts fail closed unless --auto-approve-tool-calls is enabled, and that option clicks Allow once, never a permanent grant.

Keep your setup safe and diagnose problems

Use Activity in the launcher for safe local diagnostics, and Settings then Run doctor for an end-to-end health check. Settings can also cancel a retained browser turn or remove the Codex integration before uninstall.

Temporary Chat is on by default and keeps new task conversations out of ChatGPT history. Turn it off to use ordinary saved chats, or if your connector is unavailable in that mode. Browser state is a sensitive login artifact, and the local loopback listener can be reached by processes running as the same user, so never share the launcher profile.

Read the architecture and security model in the project docs before enabling full mode, and report vulnerabilities through SECURITY.md.

What people use it for

Using ChatGPT Web models such as GPT-6 Sol inside Codex's model picker while keeping the same task interface.
Asking a ChatGPT model to read and change files in the current Codex task through Full harness mode, with approvals.
Keeping Codex quota free for other work by sending messages through your ChatGPT account instead.
Using Zero Risk mode when you want to send every prompt yourself and avoid automated page control.

Frequently asked questions

Is codex-chatgpt-web free?

The project is MIT licensed, so you can use and modify the code under those terms. Messages still count against your ChatGPT account's limits, and the README does not list prices, so check your own ChatGPT plan for costs.

What do I need to install codex-chatgpt-web?

The README lists no separate prerequisites because the app includes its browser and runtime. You need a ChatGPT account, and the release targets are macOS 13+, Windows x64 and Linux x64.

Is codex-chatgpt-web an official OpenAI tool?

No. The README describes it as unofficial browser automation, not an OpenAI API. UI changes on ChatGPT's side can break it, and drift is reported as an error rather than silently switching models.

Why don't I see Pro models in Codex?

Pro entries appear only when your account exposes them, and availability also depends on your account and workspace. Pro versions have separate entries in the model picker.

Is it safe to use on a shared computer?

The README says no. Browser state is a sensitive login artifact, and the loopback listener is reachable by processes running as the same local user, so use it only on a trusted personal workstation.

What should I use instead if I need a stable API?

The README says this project is not a supported, stable API. If you need one, use an official OpenAI API rather than this browser-based integration; the source does not name specific alternatives.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use codex-chatgpt-web if you already pay for ChatGPT, want its web models inside Codex, and are comfortable with an unofficial browser integration on a personal machine you trust. If you need a supported, stable API, work on shared computers, or run platforms outside macOS 13+, Windows x64 and Linux x64, choose another option.

Install options & downloads