Sentry's Agent Tracing is now GA. If you missed the beta: Agent Tracing captures every step of an agent run — model calls, tool executions, sub-agent handoffs, token usage, and cost — as spans inside your existing Sentry traces. Adding conversation IDs allows you to view the detailed message exchange between your agents and your users.

What makes Sentry's agent tracing different: it's failure-first and trace-connected. A bad agent response usually isn't the model's fault. It's a tool that returned the wrong shape, an argument that violated the schema, a retry loop, or a slow database query multiple layers down. Agent spans sit in the same hierarchy as your HTTP, DB, and queue spans, so when a run goes wrong you get the connected trace, log, or error to find out why.

We've also added several new features leading up to this launch:

• Conversation titles: The overview now shows a readable title for each conversation instead of a raw conversation ID, so you can spot the session you're looking for without opening it.
• Overview chart: Cost, conversation count, and total messages over time, right on the overview — so you can see usage and spend trends before you drill into a single run.
• Improved timeline and transcript views: Read a conversation the way your user experienced it — tool calls inline with input previews and durations so bad steps stand out, and clicking any message or tool call opens a side-panel with all the details.

Get started by going to Explore > Agents or check out docs for supported frameworks and platform-specific setup.