Connect ChatGPT Web models to Codex work without consuming Codex allocation
If you want to avoid stopping due to model limits during Codex work, you can utilize the separate usage limits of ChatGPT Web. This tool adds ChatGPT Web models to the Codex model selection list, allowing you to connect ChatGPT Pro models to Codex work while maintaining the existing interface, image, and streaming features. Codex allocation is not consumed, and you can configure ChatGPT to directly handle the files and terminal of the current work.
Model selection based on account permissions
The launcher detects the user's ChatGPT account permissions and automatically displays the available models. Accounts without reasoning control features are provided with Luna and Think models, while accounts with reasoning control display options from Instant to High. Extra High and Pro models appear as separate items only if the feature is enabled in the account. Each model is labeled as (Web) within Codex, and the Pro version is managed as an independent item to maintain the context budget. Sol reasoning is selected via the Effort setting, and the Zero Risk model uses a dedicated item.
Connecting work files and terminal to ChatGPT
Full harness mode is structured to return ChatGPT tool calls to the current Codex work.
- Browser-only mode Message sending is automated, but local Codex tool usage is not supported.
- Full harness mode Local Codex tool usage is possible via MCP, and automated message sending is supported.
- Zero risk mode It does not read or manipulate the ChatGPT page; prompts must be manually pasted and sent.
Self-contained runtime and installation process
The app includes its own browser and runtime, so it can run without separate installations of Chrome, Node, or Bun. It supports macOS 13 or later (arm64/x64), Windows x64, and Linux x64, and the installation script automatically selects the platform and architecture. The launcher must be terminated before updating, and the installer verifies the published SHA-256 manifest to preserve ChatGPT profiles and launcher settings. Builds do not yet have platform signatures applied, so Gatekeeper or SmartScreen warnings may appear. To run from source, Bun 1.4.0 is required, and the app can be opened with the bun run app command.
Limitations of unofficial automation and security notes
This tool is unofficial browser automation software, not the OpenAI API. Selectors may break when the ChatGPT UI changes, in which case it fails explicitly without silently changing the model or sending method. Browser state contains sensitive login information, so due to the nature of the loopback listener accessible by local user processes, do not share the launcher profile and use it only on trusted workstations. You must read the architecture and security model before enabling full mode, and vulnerabilities should be reported via SECURITY.md. This software is unrelated to OpenAI and should be used only in accordance with your own account and relevant service terms.