OpenAI Agents Python SDK v0.19.1 Fixes Guardrails and Streaming Bugs

Full analysis

OpenAI released v0.19.1 of its Python Agents SDK — the open-source framework for building agentic systems — with a batch of bug fixes targeting reliability in agent runs. Key fixes include properly canceling parallel input-guardrail tasks (safety checks that run before the model responds) when a model turn fails, canceling streamed model sessions when input guardrails fail, and correctly counting streamed retries when terminal usage data is missing. Additional fixes address SQLite-based memory session counting, WebSocket retry behavior, and tracing batch-processor enforcement on shutdown.