Hey Product Hunt 👋 I’m launching Tracebox — a black box recorder for AI agents. After debugging enough “the agent did something weird” moments, I wanted something closer to a flight recorder than a chat log: what did it see, what did it decide, which tools ran, and where two versions diverged? Tracebox records runs as append-only event streams, lets you replay and fork them, and finds the first divergence between trajectories. There’s a JS SDK, Python SDK, REST API, and an MCP server so you can inspect runs from Cursor/Claude. Try it here: • Live app: https://tracebox-ochre.vercel.app • Instant demo (no signup): https://tracebox-ochre.vercel.ap... • Docs: https://tracebox-ochre.vercel.ap... This is an early MVP. I’d love feedback from anyone building agents — especially on the replay/compare UX and what events you wish were first-class. Thanks for taking a look!

Tracebox

See exactly what your AI agent saw.

See exactly what your AI agent saw.