WebSocket support for OpenAI Responses API live on AI Gateway
Vercel’s gateway now supports persistent WebSocket sessions for the Responses API, reducing repeated context transfer during long, tool-heavy agent runs.
AI Gateway now exposes the OpenAI Responses API over **WebSocket** at GET /v1/responses. Each new turn can send fresh input plus **previous_response_id** instead of resubmitting the full context over HTTP.
Consider the persistent route for long coding-agent loops where connection and context overhead accumulate. OpenAI reports **up to about 40% faster** end-to-end execution for rollouts with at least 20 tool calls.
AI Gateway now exposes the OpenAI Responses API over **WebSocket** at GET /v1/responses. Each new turn can send fresh input plus **previous_response_id** instead of resubmitting the full context over HTTP. Consider the persistent route for long coding-agent loops where connection and context overhead accumulate. OpenAI reports **up to about 40% faster** end-to-end execution for rollouts with at least 20 tool calls. That performance figure applies to a specific tool-heavy workload, not every interaction. The route accepts raw response.create frames and supports **store=false and Zero Data Retention**, so clients must manage the socket and event protocol.
This adds transport choice, not another model-routing option: persistent WebSockets can reduce repeated connection and context overhead in long tool-heavy loops, while leaving clients responsible for socket and event handling. The reported speedup narrows the likely payoff to extended rollouts rather than ordinary chats, and store=false/ZDR preserve a path for stricter retention requirements.