OpenAI added WebSockets support to the Responses API for agentic workflows.
Why it matters
Agent loops are chatty. A coding agent or desktop agent may:
- ask the model what to do next
- run a tool
- send tool output back to the model
- repeat dozens of times
Traditional request/response calls add overhead at every step. A persistent WebSocket connection reduces that back-and-forth cost.
Reported impact
OpenAI says OpenAI models in Cursor became up to 30% faster with the new WebSockets approach.
That matters because speed is now a user-facing quality dimension for coding agents. A model that is slightly smarter but noticeably slower can still lose daily-driver status.
The architectural point is broader than Cursor. Tool-heavy agents often spend meaningful time waiting on round trips, especially when they stream intermediate reasoning, invoke tools, and resume generation many times in a single task. A persistent connection gives agent builders a cleaner transport for that loop.
Who should care
- Agent framework builders.
- Coding tools like Cursor, Codex-style IDE agents, and autonomous QA systems.
- Teams with high-frequency tool-calling workflows.
Buyer caveat
WebSockets reduce overhead, but they do not solve orchestration quality by themselves. Teams still need good tool schemas, bounded permissions, retry behavior, observability, and cost controls. The feature is best understood as infrastructure that makes well-designed agents feel faster, not as a shortcut around agent design.
Related
Sources
Primary and corroborating references used for this news item.
Spotted an error or want to share your experience with OpenAI adds WebSockets to the Responses API to speed up agent loops?
Every tool page is re-verified on a recurring cycle, and corrections land faster when readers flag them directly. If you spot a stale fact, a missing capability, or have used OpenAI adds WebSockets to the Responses API to speed up agent loops and want to share what worked or didn't, the editorial desk reviews every message sent through this form.
Email editorial@aipedia.wiki