Connect ChatGPT Pro models to Codex without consuming its quota
Imagine seeing ChatGPT Web models appear directly in the model selection list while working in Codex. Because messages are sent through your logged-in ChatGPT session, Codex's separate quota is not consumed. This tool is a launcher that uses browser automation to connect ChatGPT Web models to Codex's native model picker.
Model configuration by account tier
After installation, logging into ChatGPT in the built-in browser registers models appropriate for your account tier in Codex. Free or Go accounts can use Luna and Think models, while paid accounts can use GPT-6 Sol and GPT-5.6 Sol. The Pro version is provided as a separate item, and inference strength is adjusted via the Effort setting. Existing GPT-5.6 chats remain on that version, and model availability varies depending on account and workspace settings.
Full harness mode and tool integration
Full harness mode connects ChatGPT's tool calls to the current Codex task's files, terminal, and approval procedures via MCP. Conversations are bound to the Codex task, allowing work to continue even as context grows. The tunnel connects only externally, so it does not expose a public IP or open inbound ports. It uses the system proxy, with explicitly set proxy environment variables taking precedence. Write and modify operations are possible only if the ChatGPT workspace and administrator policies allow them.
Context limits and security settings
Context limits vary by account type and selected Effort. Plus Medium/High uses a 90,000-token window, while GPT-5.6 supports up to 270,000 tokens when experimental 3x context is enabled. GPT-6 Sol supports up to 120,000 tokens on Plus and 240,000 tokens on Pro, with compression occurring at 110,000 and 220,000 tokens respectively. Zero Risk mode does not read or manipulate the ChatGPT page; users must manually paste prompts and confirm sending. Browser state contains sensitive login information, so do not share the launcher profile and use it only on trusted workstations.
Pre-use checks
This tool is unofficial browser automation, not the OpenAI API. Selectors may break due to ChatGPT UI changes, in which case it fails explicitly rather than silently changing the model or sending method. Release packages target macOS 13 or later (arm64/x64), Windows x64, and Linux x64. Builds are not yet platform-signed, so Gatekeeper or SmartScreen warnings may appear. The installer verifies the published SHA-256 manifest before installation. This software is unrelated to OpenAI, must be used only with your own account, and must comply with applicable terms of service and workspace policies.