pypi / crewai
crewai
What changed in crewai 1, from its official release notes, upgrade guides and advisories. Also covers crewai-tools. Sources checked 2026-10-18.
3 new APIs and options2 behavior and defaults2 tooling17 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli initcrewai 124 changes
New APIs and options 3
deepinfra is now available as an OpenAI-compatible provider.
Evaluations now record why they stopped before reaching a verdict.
XPU is now available in the OpenCLIP provider's device options.
Behavior and default changes 2
Uses the LiteLLM catalog for context windows.
The run app now records what Deploy met before an attempt.
Tooling 2
Adds a command to report where a stopped model broke.
The command now tracks cost and time per run without repeated progress lines.
Fixes and other changes 17
JSON-schema models no longer warn about fields that shadow BaseModel.
Emits a call completed event for streamed Anthropic tool calls.
Correctly finds the package from pyproject.
Fixes deploy remove and lockfile checks.
Shows reasons for deployment failures.
Allows image input_files for OpenAI GPT-6 Sol/Luna/Astra.
Preserves native tool aborts.
Preserves post-hook cache identity.
Flows now learn from resumed human feedback and preserve reviewed output types.
Reads AMP's source_type during deploy push so ZIP deployments use the new code.
Trigger list and payload requests now time out after 60 seconds.
RAG now decodes local JSON files with a UTF-8 BOM.
Warns when non-empty apps resolve to zero platform tools.
GitHub loader source attribution is preserved.
SQLite flow initialization contention is fixed.
stopSequences are no longer sent to OpenAI GPT-6, GPT-5.6, or gpt-oss.
Task output display now shows the actual output.