miuuyy/codex-chatgpt-web

codex‑chatgpt‑web – Bridge ChatGPT Web into the native Codex model picker

What it does

• Provides a cross‑platform desktop launcher (macOS, Windows, Linux) that embeds a private Chromium browser.
• Logs you into ChatGPT Web inside that browser and exposes the selected ChatGPT model (Luna, Instant‑High, Pro, etc.) as a native Codex model.
• Keeps Codex’s original task UI, context lifecycle, streaming, tracing and tool‑harness intact while routing the model’s responses through a temporary ChatGPT chat.
• Two operating modes:
  1. Browser‑only – only the ChatGPT UI is used; no Codex tools are reachable.
  2. Full harness – via OpenAI’s tunnel‑client the ChatGPT model can call back into the current Codex task (filesystem, shell, image handling, approvals, etc.).

Why it matters

• Lets users who already work in Codex (OpenAI’s internal IDE) switch to the latest ChatGPT Web models without needing an OpenAI API key.
• Preserves all of Codex’s built‑in features (task compaction, tool tracing, UI) while giving access to the richer reasoning levels that ChatGPT Web offers.
• All integration is done locally; the only external traffic is the normal ChatGPT Web request and, in full‑harness mode, an outbound tunnel to OpenAI’s MCP service.

Quick start

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

# Windows (PowerShell)
irm https://github.com/miuuyy/codex-chatgpt-web/releases/latest/download/install-launcher.ps1 | iex

1. Launch the app, sign‑in to ChatGPT in the embedded browser.
2. Run the built‑in smoke test.
3. Press Install models, restart Codex, and select a ChatGPT Web — … model from Codex’s model picker.

Running from source (requires Bun 1.4.0)

git clone https://github.com/miuuyy/codex-chatgpt-web.git && cd codex-chatgpt-web
bun run app   # launches the Electron app

Full‑harness setup (optional, for tool access)

1. In the launcher open MCP, create a new OpenAI tunnel and an API key.
2. In ChatGPT settings enable Developer Mode and create a new connector named Codex Native2 with Allow all actions.
3. Paste the tunnel ID and API key into the launcher and click Connect harness.
4. Verify the runtime; the app will refuse to run with an older Codex Native connector.

Key limitations & security notes

• This is browser automation, not a direct OpenAI API. UI changes in ChatGPT can break the selectors; the bridge will error out rather than silently fall back.
• Browser cookies and login state are stored locally; anyone with access to your user account could reuse the session.
• The tunnel client only makes outbound connections; no inbound ports are opened.
• Installers are not code‑signed yet, so macOS Gatekeeper or Windows SmartScreen may warn.
• The project is unofficial and not endorsed by OpenAI; you must comply with OpenAI’s terms and your workspace policies.

License

• MIT License (see LICENSE).

Further reading

• Architecture details: docs/architecture.md
• Security model: docs/security-model.md
• Release validation procedures: docs/release-validation.md

---

This summary is based solely on the repository’s README.