What actually happened
when your agent ran?
Start with 10 or more traces to explore recurring risks, affected workflows, and where to focus fixes. A single trace offers a narrow snapshot.
Upload your traces
Recommended: at least 10 distinct traces. Include different intents, outcomes, and agent paths. More representative traces make recurring patterns easier to spot; volume alone does not guarantee coverage.
Drop your trace JSON files here
Multiple JSON files or one JSON batch · up to 10 MB per file · 50 MB total
This preview checks JSON syntax only. Format validation and Python evaluation come after layout approval.
What should the JSON look like?
Upload individual trace objects, a JSON array, or a batch object with a traces array. Each trace needs its own call context and timestamped timeline. Download the illustrative 10-trace batch below; replace every synthetic example with a distinct real trace. Final schema checks are deferred.
What to include
Context: industry, task, call ID, duration, and agents visited.
Messages: timestamp, speaker, agent name, and text.
Tool events: timestamp, tool name, arguments, and result.
Handoffs: timestamp, source agent, and destination agent.
A completed call.
One potential legal risk.
What happened
Overall risk: lowSee the evidence. Know what to fix.
Verify your work email to unlock the complete analysis: recurring patterns by agent and workflow, evidence for every finding, jurisdiction and severity breakdowns, liability context, prioritized fixes, and exports.
Free with a verified, approved work email. Verification and approval are simulated in this prototype.
Take the review with you
Original supplied evaluator output.