library · inferred from evidence
Agent
A class or component from CrewAI used in code snippets.
- Julep dev mode keeps strict diagnostics; Temporal stays prod-strict
Use deployment.prod_gap_summary() to see exactly what strict prod would reject, or result.prod_gap on a run result.
- Gemini causes 'Event loop is closed' when running inside an async context
'Event loop is closed' when running pydantic-ai's Gemini model inside an async context (e.g. Streamlit) was fixed in 0.0.46 after a regression: the maintainer first fixed it in 0.0.29, broke it again, reverted, and released 0.0.46 with the
- could not parse response
If browser-use throws 'could not parse response' against a local model, upgrade browser-use to 0.1.30 and update Ollama. The old response handling mis-parsed output from local Qwen models, which one reporter confirmed was the cause.
- Not able of making tool calls and response format at the same time
Use this workaround on the Agno model config. It forces JSON text output that Agno then parses into your schema: supports_native_structured_outputs=False supports_json_schema_outputs=True use_json_mode=False structured_outputs=True respons
- [BUG] CrewAI >= 1.4.0 - Structured outputs no longer working with azure models
CrewAI: Contributor Vidit-Ostwal patched the regression in PR #3910. Reporter-confirmed workaround: declare the structured-output Pydantic model with extra='forbid', e.g.