Rohan Paul@rohanpaul_ai · XLatitude 发布 MIT 许可的开源监控平台，将 AI 智能体对话转为生产调试数据。多数智能体已比员工接触更多用户，但对话仅存为原始日志，导致团队错过用户不满、未满足请求、重复失败和流失信号。平台提供 see, catch, fix 循环：查看会话、用户、工具、成本、延迟和 behaviors；通过 Signals 捕获重复失败；通过 MCP 从编辑器直接修复。平台面向生产智能体，关注工具使用、用户意图、重试、成本、延迟等，而非仅模型调用。推文称智能体对话是公司最被低估的数据源，Latitude 正改变此局面。
Agents token burn needs more visibility.
Latitude just launched an open-source， MIT licensed monitoring platform that turns AI agent conversations into production debugging data.
Most agents already talk to more users than any teammate， but those chats usually sit as raw logs， so teams miss frustration， unmet requests， repeated failures， and churn signals.
Latitude organizes that mess into a see， catch， fix loop： see sessions， users， tools， cost， latency， and behaviors； catch repeated failures through Signals； fix them from your editor through MCP.
The product is built for production agents， not just model calls， because agent quality depends on tool use， user intent， retries， cost， latency， and the gap between what the user expected and what the system did.