See Every AI Agent Conversation in Sentry (Beta)
Your AI agent just did something unexpected. Was it a hallucination? A bad tool call? Or did it actually handle things correctly? Sentry Conversations lets you replay the full exchange and find out.
AI conversations are multi-turn, tool-heavy, and unpredictable. Traces and logs only tell part of the story. Conversations groups everything into one scrollable view: user messages, agent responses, tool calls, token costs, and agent metadata. It's built on top of Agent Monitoring, so you keep all the data you're used to, but now the whole conversation is stitched together so you can follow the agent's reasoning step by step.
Pro tip: instrument your agents with gen_ai.conversation.id from OpenTelemetry semantics. Have your agent report the conversation ID in its telemetry (e.g. in a footnote or log), and you can query those conversations from the Sentry MCP server, or even have an agent debug itself.
Get started in 3 steps:
- Update to the latest Sentry SDK for your platform
- Set streamGenAiSpans: true in your Sentry init
- Call Sentry.setConversationId() to group conversation turns
Conversations is in beta. Try it out and let us know what you think!
Sentry Docs: https://docs.sentry.io
Conversations: https://docs.sentry.io/ai/monitoring/conversations/
#SentryIO #AIMonitoring #LLMObservability #AgentDebugging #ConversationTracing