See what every
release actually did.
Every chart knows which flag changed and when. Spot the regression, click the marker, dial it back.
p95 latency · checkout
Last 6 hLive releases
- new-checkoutrolled back25%
- search-ranker-v3healthy50%
- agent-prompt-v15healthy5%
- pricing-page-bcomplete100%
- onboarding-tourwatching10%
Watch a regression
dial itself back.
Two minutes, one bad deploy: the latency spike, the marker on the chart, the automatic rollback and the all-clear.
Three questions every
release review asks.
Which flag caused it?
Every metric is sliced by flag and variant. The answer is one click away, not one meeting away.
errors / 1k requests · by variant
exposure at detection
caught at 5% · 412 users affected · rolled back in 38 s
Did we catch it early?
Regressions show up at 5% of traffic, when a few hundred people see them — not a few million.
What should we do now?
Alerts arrive with the fix attached. Dial back, pause the experiment or page the owner — from Slack or the dashboard.
Rheostat vs. the flags
you built in-house.
| Capability | Rheostat | In-house |
|---|---|---|
| Metrics sliced by flag and variant | Yes | No |
| Automatic rollback on a rule | Yes | No |
| Percentage rollouts with schedules | Yes | Yes |
| Agent traces linked to releases | Yes | No |
| Audit log of every change | Yes | Yes |
| Evaluation in under 20 ms at the edge | Yes | No |
On-call got quieter.
“The release marker on the latency chart ended every “was it the deploy?” thread before it started.”
“We cut our mean time to recover from 47 minutes to 11. Most of that is the rollback nobody had to press.”
“One dashboard for flags, errors and latency. Our incident channel went from daily to monthly.”
Questions about metrics.
- Which metrics sources work?
- Prometheus, Datadog, New Relic, Honeycomb and anything that speaks OpenTelemetry. Custom metrics arrive over a webhook.
- Do you store my metrics?
- We keep 30 days of release-scoped aggregates. Raw data stays in your own observability stack.
- How fast does a rollback fire?
- Rules evaluate every ten seconds at the edge. Most rollbacks complete in under ninety seconds.
- Can I tune the thresholds?
- Every rule has its own threshold, window and minimum traffic, per flag and per environment.
Stop reading dashboards
after the incident.
- −61%
- median time to recover across 1,900 teams in 2026