Trace Viewer

Upload agent trace files to visualize step-by-step screenshots, actions, planner reasoning, and evaluation results.

Drop a .jsonl trace file here

Supports trajectory JSONL (PlannerTrajectoryLogger) and full-eval JSONL (run_full_eval.py)

Step through screenshots with click markers, type highlights, and scroll indicators overlaid exactly where the agent acted.

See exactly what the planner thought at every step -- its decision, instruction, reasoning, and target description.

Track partial credit with per-milestone pass/fail indicators. Instantly see where runs stall and which milestones are hardest.