Most engineering workflows separate a local task harness (Codex) from ChatGPT’s web UI, forcing context juggling and manual copy/paste. This project collapses that gap: it embeds ChatGPT Web as a selectable, native model inside Codex so each ChatGPT Temporary Chat receives the accumulated Codex task context, attachments, and images while streaming reasoning and tool activity back into the same task.
What Sets It Apart
- Native model picker integration: ChatGPT Web modes (Instant→Extra High and Pro when available) appear in Codex’s model selector, so switching to a web model is a single click while keeping Codex task lifecycle and streaming intact — browser chats are temporary and task-bound. This means you can use ChatGPT’s UI modes without losing local task metadata or breaking your Codex workflow.
- Full-harness tool routing via MCP: an optional outbound OpenAI tunnel plus a ChatGPT developer connector named for the launcher lets ChatGPT-initiated tool calls hit the current Codex task (filesystem, shell, images, configured tools). In practice this lets browser-side ChatGPT request and receive live results from your local task environment instead of only simulating actions.
- Visible, auditable flow: every browser turn is proxied into a temporary ChatGPT chat and streamed back into the Codex task (messages, visible reasoning, Markdown, and attachments), which preserves traceability for debugging and review.
Who it's for and tradeoffs
Great fit if you want to combine ChatGPT Web's UI/modes with a local Codex task harness — for interactive debugging, coding with local tools, or hybrid workflows that need local files and terminal access. It is also useful when you prefer ChatGPT Temporary Chats (privacy mode) but want them tied to a reproducible, local task context.
Look elsewhere if you need a supported, long-lived API integration or cannot sign in to a ChatGPT account: this is unofficial browser automation (not an OpenAI API), the bridge depends on browser UI selectors and ChatGPT developer-mode connectors, and UI changes can break functionality. Browser state is sensitive (login artifacts), so use it on trusted machines and with your own account. The Pro mode on accounts may have capability differences and cannot initiate local MCP/tool calls in some configurations.
