v2.65.0

New Features

  • Added PayloadSelfWritePaths and PayloadSelfReadPaths fields to AIAgent table for granular control over agent's self-execution payload access
  • Implemented operation-level payload control for AI agents with flexible API key management
  • Added SQL logging pattern filtering with simple wildcard support for better debugging control

Improvements

  • Enhanced AI agent payload management with upstream/downstream path controls for fine-grained state modification
  • Improved validation handling for base agent and loop agent types
  • Made metadata checking optional for agent config loading when driver class exists
  • Added support for consistent newPayload/previousPayload handling in agent execution chain
  • Made database schema generation failures non-fatal to improve resilience

Bug Fixes

  • Fixed MaxRows parameter being ignored in RunViewGenericInternal
  • Fixed variable scoping issues in catch blocks for better error handling
  • Added safeguards to prevent failed agent steps from breaking the payload chain
  • Fixed type inconsistencies in payload handling between agent execution steps