CrewAI 1.15.21
CrewAI 1.15.21 introduces telemetry for checkpoint runtime and CLI usage, alongside multiple provider bug fixes, UTF-8 checkpoint persistence, and corrections to gpt-4o-mini context limits.
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 version 1.15.21 introduces checkpoint runtime and CLI usage telemetry. Key bug fixes include handling gateway errors returned within HTTP 200 responses, maintaining deploy push on AMP create source, gracefully reporting scrape failures instead of raising IndexError in Oxylabs, routing DashScope models through their native provider, saving JSON checkpoints as UTF-8, and removing extra '$' characters from BrightData search URLs. In schema handling and models, list-form type arrays are now supported in JSON schema conversions, and the gpt-4o-mini context window has been corrected to 128,000 tokens (down from 200,000). Streaming tool call arguments are now preserved at contentBlockStop, and pre-commit hooks have been made portable for Windows. Documentation updates address telemetry vs. tracing separation, streaming output examples, and skill install references.
可执行摘要
CrewAI v1.15.21 adds checkpoint runtime and CLI telemetry, fixes HTTP 200 gateway error handling, preserves streaming tool call arguments at contentBlockStop, routes DashScope via native provider, corrects gpt-4o-mini context window to 128000, and ensures JSON checkpoints save as UTF-8.
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Verify gpt-4o-mini context window usage to ensure prompts remain within the corrected 128k token limit.
- Update workflow integrations that rely on Oxylabs error handling, BrightData search URLs, or DashScope model routing.