New Release - 0.157.0

Announcements
1

Core Improvements & Fixes

  • Stronger Agent Autonomy

    • Agents can now act with more independence, thanks to new internal capabilities and clearer context management in planning.

  • Planning Cache Introduced

    • The new cache system significantly accelerates repeat agent runs (especially for long workflows). Community reports confirm faster iterations after the first execution, but note the first run is slower. Issue #991

  • Tool API Overhaul: Required description, Typed Returns

    • All Tools must now define a non-empty description and return a ToolResult. This enforces uniform documentation and output—but breaks older plugins/tools that don’t comply. Issue #996

  • Agent.run_sync Removed

    • Synchronous agent runs must migrate to async (see docs). Confusion around this was quickly addressed, but check your deployment scripts. Issue #995

  • Improved Error Handling

    • Crew/agent runs fail more gracefully when tool execution throws exceptions; this closes a top user complaint from v0.156.x. Issue #990

  • API Naming Harmonisation

    • The agent context API and naming is now more consistent, lowering the learning curve for new contributors and integrators.

New Features & Enhancements

  • Better Crew Run Safeguards

    • Enhanced validation and clearer error reporting when launching or managing agents/crews.

  • Interface Consistency

    • Developers will notice improved naming standards and type uniformity when building tools or custom agents.

  • Faster CI/CD and Repeated Planning

    • The planning cache is designed for teams that re-run similar workflows, making test/production cycles more efficient—after the initial run.
1 Like