Herm gives you two observability surfaces:
- The Herm web app for tracking sessions, inspecting debug events, and
understanding cost and usage across your workspace.
- The Sessions API for exporting the same session events and usage
snapshots into tools such as Langfuse, Datadog, Honeycomb, or your internal
logging pipeline.
Track sessions in the web UI
Open Sessions in the Herm app to see the runs created by your agents. Each
session is tied to an agent, a subject, lifecycle timestamps, transcript
history, and usage data when available.
Use this view when you want to review what happened across all customer-facing
or internal agent sessions without building your own admin console first.
Debug a session
Every session detail page includes a Debug view. Use it to inspect the raw
event stream behind a conversation, including user messages, agent messages,
reasoning, tool calls, tool results, approvals, status changes, and usage
snapshots.
The debug view is useful when you need to answer questions such as:
- Which tool did the agent call?
- What result came back from that tool?
- Did the turn stop because it finished, needed approval, or required a custom
tool result?
- How much context, latency, and estimated cost did the turn use?
Analyze cost and usage
Session details show cost and usage data when Herm has a usage snapshot for the
run. Usage can include tokens, context window percentage, model calls, latency,
estimated cost, and cost status.
For the raw usage shape, see
Usage snapshots.
Use the dashboard
The Herm dashboard aggregates workspace-level activity so you can monitor agent
usage over time. Use it for a quick read on active sessions, message volume,
agent activity, and usage trends before drilling into an individual session.
Export to another provider
If your team already uses an observability product, stream Herm events from your
backend and forward them into that system.
Use:
For example, you can forward terminal turn events with usage into Langfuse for
per-generation token, cost, model, and latency reporting. You can also forward
tool-call events into Datadog, Honeycomb, or your own logs to reconstruct each
agent run end to end.
Keep the Herm sessionId, turnId, event id, and event sequence on any
external record you create. Those fields let you jump back from your external
observability tool to the source session in Herm. Last modified on July 29, 2026