Roll out without the dread
Move a release from 1% to 100% on a schedule you set, with error rate, latency and business metrics checked at each step. If something moves the wrong way, Kerf stops.
A rollout plan you can read at a glance
- 01
Staff first
Turn the change on for your own team for a day. Real usage, no customer risk.
- 02
1%, then 5%
A small slice of production. Enough traffic to catch crashes, small enough to contain them.
- 03
25%, then 50%
Kerf compares the exposed group with the rest on your guardrail metrics before each increase.
- 04
Everyone
At 100% the flag is marked ready for cleanup and the experiment, if any, is closed.
Health checks that actually gate the next step
Attach any metric to a rollout step. Error rate from Datadog, p95 latency from your APM, checkout conversion from the warehouse. A step only advances when every check passes for the window you chose.
- Hold, advance or roll back automatically
- Different thresholds per step
- Page the owner in Slack when a check fails
Built for the awkward cases
Ring based release
Roll out by account tier, region or tenant, so your biggest customers go last.
Scheduled steps
Advance at 10am on weekdays only, never on a Friday, never during a freeze.
Approval gates
Require a named approver before a step touches a regulated segment.
Instant rollback
One click reverts to the previous state. SDKs pick it up within a second over streaming.
Mobile aware
App store releases lag, so rollouts target by app version and fall back safely for old builds.
Full history
Every step, check result and approval is logged and exportable for audit.

