Instrument your first trace
Drop in the SDK and run a pre-built retrieval agent end to end.
Let's get started
Welcome to Kestrel
2/4
You've run an agent in Workbench
You've explored the Prompt Lab
Score a live evaluation
Measure answer quality and catch regressions before you ship.
Route traffic to a monitor
Alert the on-call channel when latency drifts past your budget.
New in Kestrel: Flightpath Build multi-step agents and ship them without writing glue code. Take a look
Tracing
| Name | Most Recent Run | Trace Count | Error Rate | P50 Latency | P99 Latency | Description |
|---|---|---|---|---|---|---|
| checkout-copilot | a day ago | 7 | 0.8% | 2.94s | 4.16s |
Datasets & Experiments
| Name | Description | Experiments | Last Experiment | Examples | Type | Updated At | Created At |
|---|---|---|---|---|---|---|---|
| Refund edge cases | 3 | 22 minutes ago | 41 | kv | a day ago | 6 days ago |
Prompts
| Name | Visibility | Prompt Type | Created By | Last Updated | Commits | Last Commit | Commit Tags |
|---|---|---|---|---|---|---|---|
| support_triage_v3 | Private | ChatTemplate | dana.reyes+kestrel@northreef.io | an hour ago | 4 | b71c4ae9 |
Annotation Queues
| Name | Total | Type | Assigned Reviewers | Updated At | Created At | Description |
|---|---|---|---|---|---|---|
| Escalation spot-checks | 12 | Manual | 2 | 3 hours ago | 5 days ago | Weekly sample of handoffs to a human agent |
| Tone & policy review | 29 | Sampled | 4 | yesterday | 11 days ago | Flags refusals and off-policy phrasing |