Comments
Collaborate directly on traces without leaving the context.
When something looks wrong in a trace, the conversation about it usually ends up in Slack or a meeting — disconnected from the actual data. Comments bring that conversation into the trace itself, pinned to the exact span where the issue is.
Engineers, PMs, and domain experts can all annotate the same run. Everyone sees the same data, and the discussion stays attached to it permanently.
What you can do
Highlight and reference anything in a span — a specific part of the prompt, a tool output, a token count — and attach a comment directly to it. The highlight is preserved so anyone reading later knows exactly what you're pointing at.
Tag teammates with @mention to pull them into the conversation. They get notified and can respond in context, without needing to find the trace themselves.
React to comments to acknowledge, agree, or flag without adding noise to the thread.
Keep the conversation in one place. Instead of screenshotting a span and pasting it into Slack, link the trace and comment on it directly. The full context — every span, every input and output — is right there for whoever you tag.
Adding a comment
Open any trace, click a span, and use the comment panel on the right. Comments are visible to everyone on your team and persist across sessions.
Why it matters
Most agent issues aren't obvious from code alone. A PM notices the response tone was off. A domain expert sees the retrieved document was irrelevant. An engineer spots the tool returned a stale value. These are different people with different context, and they're usually not looking at the same tool at the same time.
Comments make it possible for all of them to work on the same trace, in the same place, without coordinating a meeting or losing the thread in Slack.