OpenAI Agents SDK v0.17.7 - Configurable WebSocket max_size Enhances Streaming Robustness

What changed and why it matters

OpenAI Agents SDK v0.17.7 introduces a configurable WebSocket `max_size` limit. In practice, this lets developers raise or lower the ceiling for a single message's byte size that the SDK will accept over an established WebSocket connection. For agent-driven systems that stream large tool outputs, logs, or batched messages, the default WebSocket size can be a hidden source of dropped connections or exceptions; making it configurable avoids those runtime surprises.

Business and operational impact

For organizations running production agent workflows-RAG pipelines, multi-tool agents, or real-time assistants-this change improves reliability and reduces incident noise. Teams can increase the limit when sending large model outputs or long serialized tool payloads, or reduce it to constrain memory usage and surface oversized messages early. The ability to tune `max_size` also helps integration with existing infrastructure (proxies, load balancers, client libraries) that may have different limits or buffering behavior.

What leaders should do next

Evaluate agent workloads to determine realistic message size distributions and decide on a sensible `max_size` default for your environment. Update SDK dependencies in a controlled rollout, add tests that simulate oversized payloads, and monitor connection errors and memory usage after changing the limit. Where possible, adopt defensive patterns-chunking large payloads, compressing streams, and validating message sizes before send-to balance robustness and resource efficiency.

Overall, while modest, this release demonstrates careful production hardening that reduces operational friction for teams building at scale with OpenAI agent tooling.