openai-agents-python
Facts
- Openness
- Open source
- Licence
- MIT
- Language
- Python
- Pinned version
ce8ae9f(2026-10-02)- Topics checked
- 29 of 33
- Kind of agent
- General framework
What it does notably well
- Concurrency
openai-agents-python runs its function tools as asyncio tasks, offloads sync function tools with asyncio.to_thread, and shares one httpx connection pool (shared_http_client)…
- Message model
When RunState.reject() records a rejection for a pending approval and the run resumes, openai-agents-python builds a ToolCallOutputItem through function_rejection_item whose…
openai-agents-python's drop_orphan_function_calls (run_internal/items.py) removes tool and program call items that have no output, plus reasoning items left dangling by the drop…
- Errors and failover
openai-agents-python's model-retry layer, once enabled with a policy, is bounded by max_retries, delays retries with capped jittered backoff or the provider's retry-after value…
- Eval harness
openai-agents-python pairs a public deterministic ScriptedModel (src/agents/testing/model.py) with an autouse conftest fixture that fails tests calling the OpenAI model classes…
- External memory
openai-agents-python defines a Session protocol and a SessionABC (src/agents/memory/session.py) that nine classes extend directly (SQLiteSession, AsyncSQLiteSession…
See how openai-agents-python does on each topic
The topic-by-topic verdicts, the code lines and a side-by-side comparison are in the app. Sign in with Google; new accounts are approved by hand.
Independent analysis, not affiliated with openai-agents-python.