Your ChatGPT account, dressed up as an OpenAI API

What it does ChatMock is a local server that authenticates to your ChatGPT account and exposes an OpenAI-compatible API on localhost:8000. Point any client that supports a custom base_url at it—IDEs, chat apps, or scripts—and it translates requests for the underlying Codex service. It also offers Ollama-compatible endpoints for broader tool support.

The interesting bit Rather than offering a dumb proxy, it maps Codex-specific features—configurable reasoning effort, thinking summaries, fast mode, and web search—into standard OpenAI API shapes. It even serves /v1/responses over HTTP and WebSocket, and can advertise each reasoning level as a separate model string.

Key highlights

• OpenAI-compatible /v1/chat/completions and /v1/responses (HTTP and WebSocket)
• Ollama-compatible endpoints for tools in that ecosystem
• Exposes vision, tool calling, and web search
• Configurable reasoning effort, summary verbosity, and compatibility modes via CLI flags or environment variables
• Optional fast mode for supported models

Caveats

• You log in with a ChatGPT account, not an API key, and the README warns to “use responsibly and at your own risk”
• Unofficial and unaffiliated with OpenAI, so it may break without warning if the underlying service changes
• Uses non-standard model names like gpt-5.4 that may confuse clients hardcoded for official OpenAI model strings

Verdict A pragmatic stopgap if you want to wire Codex into an IDE or chat client that only speaks the OpenAI SDK, but skip it for production workflows that need stability or official support.