OpenAI Agents SDK Python v0.21.1
OpenAI Agents SDK Python v0.21.1 adds model call timeouts, improves sandbox controls (disabling Docker networking, run-scoped working directories, Modal resource options), and addresses several core, realtime, and session-handling bugs.
Why this signal matters
OpenAI Agents SDK Python version 0.21.1 includes new features and bug fixes across core modules, sandboxes, sessions, and realtime handling. Highlights include: - Core: Implemented model call timeouts, enforced exact call approval decisions, accounted for Responses requests without usage data, ensured cleanup of MultiProvider child instances following failures, and rejected partially matched stacked anchors. - Sandboxes: Added run-scoped working directories, enabled disabling network access in Docker sandboxes, added Modal sandbox resource configurations, validated raster content for view image tools, and enforced POSIX normalization on model and patch paths. - Sessions: Resolved SQLite table conflicts for advanced structures and integrated compaction token usage into run totals. - Realtime & Chat Completions: Resolved Chat Completions reasoning replay issues, added audio truncation at zero elapsed time, and handled proper iteration termination on clean server disconnects.
Actionable summary
OpenAI Agents SDK v0.21.1 introduces core model call timeouts, run-scoped sandbox working directories, Modal sandbox resource options, and disabled networking support for Docker sandboxes. Fixes cover MultiProvider resource cleanup, call approvals, token/request tracking, realtime session termination, and SQLite schema conflicts.
- Agent usefulness
- 82/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade openai-agents Python package to version 0.21.1.
- Review sandbox configurations if utilizing Docker network restrictions or custom Modal resource allocations.
- Evaluate timeout parameters on model calls where long-running operations require explicit limits.