feature
Add spreadsheet reading functionality for agents in crewai-files
Agents can now read and process spreadsheet files, expanding data handling capabilities.
Evidence
+Add functionality to read spreadsheets for agents in crewai-files.
CrewAI
Agent frameworkswww.crewai.com19 published changes
2 sources under watch
Change timeline
feature
Add support for LiteLLM-compatible kwargs
Developers can now pass LiteLLM-compatible parameters directly, improving integration flexibility with alternative LLM providers.
Evidence
+Add support for LiteLLM-compatible kwargs.
other
Fix tool actions to honor failure policies and get fresh usage budgets per kickoff
Tool actions now properly respect failure handling policies and reset usage budgets on each crew execution, improving reliability.
Evidence
+Fix tool actions to honor failure policies and get fresh usage budgets per kickoff.
feature
New crewai eval command added to report where a stopped model broke.
Users can now diagnose model failures more effectively using a dedicated eval command.
Evidence
+Add crewai eval command to report where a stopped model broke.
api
Add deepinfra as an OpenAI-compatible provider
Users can now use Deepinfra as an alternative OpenAI-compatible LLM provider, increasing provider flexibility.
Evidence
+Add deepinfra as an OpenAI-compatible provider
other
Fix crewai eval exit codes and add login prompts for unattended tracing
Improved error handling with proper exit code 1 when evaluation gates fail, and better user guidance for authentication.
Evidence
+Ensure crewai eval exits with code 1 unless the gate passed and prompt to log in when nothing was traced unattended
feature
Support aliases as connection identifiers
Developers can now use more flexible naming for connections in CrewAI
Evidence
+Support aliases as connection identifiers
github.com · also seen at 1 more source
feature
Add Clipper integrations client
Users can now integrate with Clipper for additional functionality.
Evidence
+Add Clipper integrations client
featurehigh
Promote conversational flows to stable
Conversational flows are now stable and recommended for production use, improving API maturity.
Evidence
+Promote conversational flows to stable
feature
Add declarative conversational flows documentation and enable declarations to drive conversational mode
Developers can now use declarative syntax to control conversational flows, with conversational mode being opt-in and unmistakable.
Evidence
+Add declarative conversational flows documentation
Synthesize built-in conversational methods for declarations
Enable declarations to drive conversational mode
Make conversational opt-in unmistakable
feature
Add reporting of flow outcome, duration, and human-in-the-loop signals.
Users can now track flow execution outcomes, timing metrics, and human-in-the-loop interaction signals for better flow monitoring and debugging.
Evidence
+Report flow outcome, duration, and human-in-the-loop signals.
otherhigh
Bump torch to version 2.13.0 to address security vulnerability.
Security patch for torch dependency - users should update to get the vulnerability fix.
Evidence
+Bump torch to version 2.13.0 to address security vulnerability.
api
Fix for LiteLLM-routed models provider preservation
Developers using LiteLLM routing will have provider information correctly preserved in their models.
Evidence
+Fix preservation of provider on LiteLLM-routed models.
feature
Add project_id support to link OSS usage to enterprise accounts
Users can now link their open-source CrewAI usage to enterprise accounts for better usage tracking and management.
Evidence
+Add project_id to link OSS usage to an enterprise account
github.com · also seen at 1 more source
other
Security updates: bump aiohttp and cryptography dependencies
Closes six GHSA security advisories by updating critical dependencies, improving security posture.
Evidence
+Bump aiohttp and cryptography to clear six GHSA advisories
positioning
Messaging shifted from 'acceleration and scale' to emphasis on 'governance' and 'central governance'
Positions CrewAI as an enterprise-grade solution focused on operational control and compliance, not just speed
Evidence
−Accelerate and scale Agentic AI adoption across the organization
+Scale agentic AI across the organization, with the governance to keep it in production
other
Support model changed from fixed '50 hours of development/month' to flexible 'a la carte' engineering and training
Offers greater flexibility for enterprises with variable support needs
Evidence
−50 hours of development/month
+Forward deployed engineering and training, available a la carte
www.crewai.com · also seen at 1 more source
feature
Skills Repository promoted from experimental status
Users can now use Skills Repository as a stable, production-ready feature without experimental constraints.
Evidence
+Promote Skills Repository out of experimental status
otherhigh
Tool-result caching changed from default-on to opt-in
Users will need to explicitly enable tool-result caching; default behavior now skips caching, potentially affecting performance expectations.
Evidence
+Make tool-result caching opt-in instead of on by default