Home
Learn what Trailsense is, what telemetry it ingests and how to instrument your app correctly.
Trailsense helps teams understand AI agent behavior from OpenTelemetry traces. Rather than showing only raw span data, it organizes telemetry into readable conversation turns so you can quickly see what the user asked, what the agent answered, and how it arrived at its response.
Teams responsible for agent quality can quickly review Sessions and identify issues in agent behavior. When they find a problem, they can create an issue, assign it to a teammate, and track agent progress in one place. Previously, this process often relied on emails, screenshots, and Slack messages while information was lost along the way (if it was attached in the first place - yes we have been there 😊).
If you're reading this, you're probably an agent developer looking to integrate Trailsense into an existing product. We recommend spinning up a GitHub Codespace
or dive into the documentation:
Core Concepts
| Concept | What it means in Trailsense |
|---|---|
| Session | A group of related traces, usually tied together by gen_ai.conversation.id. |
| Trace | One end-to-end execution tree identified by a single OTEL trace ID. |
| Root span | The top-level span for one agent turn or workflow invocation. |
| Child spans | Nested spans for LLM calls, tools, and other internal work. |
| Conversation view | The UI built from the root span's trailsense.input and trailsense.output messages. |
| Trace view | The detailed span tree, timing, metadata, and issue workflow for a trace. |