Latitude

Latitude 做的是 AI Agent 的可观测性与监控：把每次运行里的提示词、工具调用和返回结果串成一条链路，出问题能直接定位到哪一步、哪个变量变了，而不是翻日志猜。除追踪还有评测与数据集管理，改完提示词拿同一批样本重跑，看分数涨了还是掉了。核心开源，可以自托管，也能用托管版本，适合已经在生产里跑多个 Agent 的团队。 🔗 文档: https://docs.latitude.so | GitHub Stars: 4.7k+

Latitude handles observability and monitoring for AI agents: prompts, tool calls and responses from each run are stitched into a single trace so a failure points at a specific step and a changed variable instead of a log hunt. Evaluations and dataset management sit alongside tracing, so a prompt change can be rerun against the same samples and scored. The core is open source and self-hostable, a hosted version exists, and it fits teams already running several agents in production. 🔗 Docs: https://docs.latitude.so | GitHub Stars: 4.7k+