How It Works
1
Connect your agent
Use the setup wizard, a framework integration or OpenTelemetry. Run the agent and check that its
messages, tool calls and outcomes arrived in Telemetry.
2
Define what to watch
Describe a failure in Signals and add examples. Prepare its evidence capture and evaluation
settings for the workflow you want to monitor.
3
Review the evidence
Open a decision to see the saved prediction and cited tool results. Record a correction when the
result is wrong, or mark the evidence as insufficient.
4
Check a change
Use the evidence to investigate the agent’s code or instructions. Compare runs after a change
and review proposed fixes before merging them.
Get Started
Send your first trace
Follow a complete Python or Node.js example and inspect its tool result.
Set up from your terminal
Use the published setup wizard and Doctor, including JSON output for coding agents.
See a Recorded Agent
Explore a concrete workflow before connecting your own application. These walkthroughs show recorded sample-agent runs, including failures and unresolved outcomes. They are examples to inspect, not live monitoring of your workspace.Insurance submission
Compare a claimed submission with its tool result and a later recovery.
Document review
Inspect missing evidence, stale document versions and a corrected review.
Support agent
Follow shipment requests, retries, queued work and confirmed outcomes.
Explore the Docs
Sessions & Tracking
Understand the messages, tools and outcomes in a run.
Issues & Diagnosis
Define a signal, inspect a prediction and review its supporting evidence.
Ingestion Options
Choose native SDK tracking or an existing OpenTelemetry pipeline.
Use with AI
Give coding agents the docs and supported investigation commands.