Hey everyone, I’m Tussa, founder of Traser. I started building Traser because I kept running into the same problem while working on AI systems myself: the trace was there, but figuring out which part of the run actually mattered still meant manually digging through everything. Talking to other engineers confirmed it wasn’t just my workflow. You end up going through tool calls, state, retrieval, retries, handoffs, and outputs trying to figure out where behavior started to diverge. Traser is my attempt at making that investigation smaller. You give it a suspicious run, optionally compare it against other runs, and it narrows the trace down to a few evidence-backed places worth checking. I’ve talked to other engineers while building it, and one of the biggest lessons was not to pretend the tool can magically determine root cause. The engineer still decides that. Traser’s job is to reduce how much of the run you have to inspect before you know where to start digging, so you don’t spend an entire afternoon debugging. It’s free to try, runs locally in your browser, and doesn’t require an SDK or production instrumentation just to get started. Would love feedback from anyone building agents or AI workflows, especially if you have a weird failed run you can throw at it. I’m just as interested in where it gets things wrong as where it helps.

Traser

Find where your workflow first went wrong.

Find where your workflow first went wrong.