Codex ChatGPT Web is a desktop launcher that adds the ChatGPT Web models on your account, including Pro, to the native model picker in Codex. Turns sent to those models count against ChatGPT Web's own usage limits rather than your Work or Codex quota, while Codex keeps its interface, tasks, images, and streaming. It is unofficial browser automation, not an OpenAI API, and is not affiliated with OpenAI.
The launcher bundles its own browser and runtime. You sign in to ChatGPT in the embedded browser, install the models into Codex, and choose an entry ending in (Web). Available models depend on the account: free plans get Luna and Think, and paid accounts get GPT-6 Sol and GPT-5.6 Sol with effort levels from Instant to High, plus Extra High and Pro where the account exposes them. Conversations stay tied to the Codex task, and native Codex compaction keeps working as context grows.
Three modes control how far the bridge reaches. Browser-only sends messages automatically without local tools. Full harness connects ChatGPT's tool calls back to the current task's files, terminal, tools, and approvals through MCP over the official OpenAI tunnel client, an outbound tunnel that needs no public IP or open inbound port. Zero Risk never operates the ChatGPT page: you paste and send a prepared prompt yourself, and tools run through a separate MCP connector. Unexpected approval prompts fail closed unless automatic approval is explicitly enabled, and interface changes on the ChatGPT side fail explicitly rather than silently switching model or transport.
Features
- Native model picker: ChatGPT Web models, including Pro versions, appear as separate entries in Codex
- Separate usage limits: turns use ChatGPT Web allowances instead of Codex quota
- Full harness mode: ChatGPT tool calls reach the Codex task's files, terminal, tools, and approvals through MCP
- Zero Risk mode: manual paste-and-send with tools through a separate MCP connector, without automating the ChatGPT page
- Outbound tunnel: uses the official OpenAI tunnel client, with no public IP or inbound port
- Larger context: experimental extended context windows on supported plans, with Codex compaction
- Subagents: a compatibility protocol for cross-backend subagents and a native mode with Web-to-Web delegation
- Diagnostics: an activity view, an end-to-end health check, and removal of the Codex integration before uninstall
- Cross-platform launcher: macOS, Windows, and Linux builds with checksum-verified installers
