Gideon (@gideonxqt)

Just look what I found, an open-source Sentry, but for AI agents. It shows what's about to break in your agent before users notice - failed traces grouped into tracked issues, evals built from your team's actual judgments, full multi-turn sessions and tool calls in one view. Every trace is semantically searchable too — find anything by meaning, zero sampling. How to try it: - sign up free at latitude, grab your API key + project slug - paste one prompt into Claude Code, Cursor, or Codex – it wires up tracing for you automatically - run your app like normal, every LLM call now shows up as a trace - open the dashboard and watch issues get grouped and tracked in real time Provider-agnostic — OpenAI, Anthropic, Bedrock, LangChain, all of it. Free tier: 20K credits/month, unlimited seats. MIT licensed, 4.1k stars. If you're shipping an agent without something watching it, you're flying blind. link below👇