We've redesigned Conversations: the view you see when you navigate from Explore → Conversations, so you can follow every message, tool call, and model response in an agent conversation.
Conversations builds on Sentry’s Agent Tracing, which automatically captures agent runs, tool calls, model interactions, and errors across the entire agent workflow.
What’s new in the lists view: A chart at the top shows high level information that makes it easier to find the chat you are looking for. Each conversation is titled now for easier identification. We've added a new data point for duration along with expanding the list of tools used in-line rather than on hover. The entire page can be filtered down from all types of chats to ones with human users vs ones that are agent to agent interactions.
You can also now save conversation lists with your preferred filters via Explore's saved queries. Save as stores your current filters (and agent).
What's new in the details view: Transcript and Timeline views
Transcript: The Transcript view shows the full back-and-forth between user, model, and tools in a chat-style layout. Tool calls appear inline with a short input preview and duration, and long runs collapse so the conversation stays scannable. Selecting any message opens a side panel with duration, metadata such as model and cost, and the full Input, output, and attributes.
Timeline: The Timeline view shows every span in the conversation in order, so you can spot latency, errors, and the sequence of tool calls. Clicking a span opens a side panel with additional high level data (cost, model) as well as its input, output, and attributes.
Conversations groups spans by gen_ai.conversation.id, so whether a user's session spanned multiple traces or a single trace contained several conversations, you get the full picture.
To get started with using Conversations in Sentry, you need to first ensure you have Agent Tracing enabled:
Learn more in our agent tracing docs.
Github
Slack