Speeding up agentic workflows with WebSockets in the Responses API
read at source ↗ openai.com
Speeding up agentic workflows with WebSockets in the Responses API
Source: OpenAI Date: 2026-04-22 URL: https://openai.com/index/speeding-up-agentic-workflows-with-websockets
Summary
OpenAI’s April 2026 technical announcement of WebSocket support in the Responses API — enabling persistent, bidirectional connections between applications and the API rather than the stateless HTTP request-response model. WebSockets allowed agentic applications to maintain live connections, receive streaming updates without polling, and handle multi-step workflows with lower latency and less connection overhead. This was a developer infrastructure improvement with significant implications for real-time agentic applications.
Implications
WebSockets as agentic infrastructure. Long-running agent tasks — crawling, multi-step coding, research gathering — that took minutes rather than seconds needed a connection model that didn’t time out or require constant reconnection. WebSockets in the Responses API made these workflows practical in production, removing a significant implementation burden from developers building complex agents.
Thread: Agentic API platform. Sits in the technical infrastructure build-out alongside the Responses API launch, MCP integration, AgentKit, and the Codex CLI as the developer-facing API improvements enabling production agentic applications. WebSocket support was a relatively unglamorous but practically important addition.
Watch: Whether WebSocket support in the Responses API was adopted as the standard connection pattern for multi-step agentic applications, and how the latency and reliability improvements compared to the previous polling-based approach in practice.