Module chatgpt 

Functions

backend_api_handler

ChatGPT backend calls outside the model rail are not model JSON and must not be compressed or cost-metered. They are credential-preserving passthroughs.

codex_responses_handler

Codex subscription model turns hit ChatGPT’s Responses-compatible rail: /backend-api/codex/responses. Forward through the same compressor/metering path as OpenAI Responses, but target https://chatgpt.com.

codex_responses_ws_handler

ChatGPT’s Codex rail rejects WS-only continuation fields such as previous_response_id; ask Codex to retry through the HTTP/SSE path.