What crewAIInc/crewAI shipped
Written by FoxPlug from public releases; not affiliated with CrewAI. An automatic summary of the public release, pull request and commit data of github.com/crewAIInc/crewAI. CrewAI did not write it and does not use or endorse FoxPlug. Every line links to the public change it describes.
Get a weekly update like this for your product, free
Week of September 21, 2026
What shipped
- Provider throttles now retry automatically inside LLM client calls with exponential backoff, without consuming agent retries or iteration counts. Pull request #7677
- New `crewai eval` command evaluates the last traced run through the evaluation service, or a specific run by ID. Pull request #7649
- Traced runs are now recorded to `.crewai/last_run.json` for evaluation instead of printing the execution ID. Pull request #7648
- Bedrock provider falls back to synchronous calls in a worker thread when the async SDK is unavailable. Pull request #7680
- CLI crew wizard now prioritizes commonly used Platform integrations like Gmail and WhatsApp. Pull request #7573
- Generated `AGENTS.md` now documents Platform tool discovery and how to use tools in Python or JSON. Pull request #7581
- Security policy now includes private vulnerability reporting on the repository and a security@crewai.com contact. Pull request #7717
- SeleniumScrapingTool now keeps the browser window open for reuse across multiple calls to the same instance. Pull request #7663
- S3ReaderTool now properly closes response bodies to release HTTP resources. Pull request #7647
Why it matters
Evaluation is now integrated into the CLI workflow, making it easier to assess crew performance. Provider resilience improvements handle transient failures gracefully. Security and resource management fixes ensure reliability and proper cleanup.
Changelog entry
- fix(llm): retry throttled provider calls with bounded exponential backoff Pull request #7677
- fix(bedrock): fall back to sync calls when aiobotocore is unavailable Pull request #7680
- feat(cli): prioritize popular platform integrations in crew wizard Pull request #7573
- feat(cli): new crewai eval command to evaluate last traced run or specific runs Pull request #7649
- feat(tracing): record last traced run to .crewai/last_run.json Pull request #7648
- docs(cli): document platform tool discovery in generated AGENTS.md Pull request #7581
- fix(tools): keep SeleniumScrapingTool browser window open for reuse Pull request #7663
- fix(tools): close S3ReaderTool response bodies to release resources Pull request #7647
- security: add private vulnerability reporting and security contact Pull request #7717
This week: automated retry logic for provider throttles, native evaluation through crewai eval, Bedrock fallback support, and better resource management across tools.
CrewAI shipped several reliability and usability improvements this week. Throttled provider calls now retry automatically without affecting agent iteration counts. The new crewai eval command lets you evaluate your last traced run directly from the CLI. Bedrock provider calls gracefully fall back to sync when async is unavailable. Plus fixes for resource cleanup in S3 operations and browser reuse in Selenium scraping.