CrewAI 1.15.29
CrewAI 1.15.29 promotes AgentExecutor from experimental status, adds Azure DocumentDB vector search, enhances evaluation CLI workflows, and fixes several memory leak and tool failure policy bugs.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
CrewAI release 1.15.29 brings several major features, fixes, and refactors across agent orchestration and evaluation. Key additions include promoting AgentExecutor out of experimental status, supporting LiteLLM-compatible kwargs, and integrating Azure DocumentDB vector search. The CLI and evaluation tooling now support offline guardrail evaluation runners, one-command crew execution for verdicts, and file exports for evaluation data. Critical bug fixes address agent stability and security: stopping agents from repeatedly issuing the same tool call, preventing private memories without sources from leaking during recall, honoring tool failure policies with fresh budgets per kickoff, and escaping scope and ID values in LanceDB filters. Context compaction was also refactored into a BaseLLM capability.
可执行摘要
CrewAI version 1.15.29 released. Key changes include promoting AgentExecutor out of experimental, adding LiteLLM-compatible kwargs, Azure DocumentDB vector search, offline guardrail evaluation runners, tool execution loops/failure policy fixes, and private memory leakage prevention.
- Agent 实用度
- 62/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade crewAI dependency to version 1.15.29.
- Review tool error handling configurations to ensure alignment with newly enforced failure policies and per-kickoff usage budgets.
- Verify LanceDB and private memory queries to ensure expected behavior given fixes to leakage and filter escaping.