Product feature
Release safety
Ramp a change on a timer and let your monitoring switch it off when a step goes bad. On Business, production can also demand a written reason for every change, or freeze outright for a sales weekend.
Last updated:
Featureflip's release-safety controls limit how far a bad flag change can spread and how long it lasts. A rollout ramp raises a percentage rollout through steps you set, so a regression reaches a few users first, and if a Datadog or Grafana alert fires, a flag trigger turns the flag off within seconds. Production gets two more guards on Business. Every change there can be made to carry a written reason, and a change freeze can stop changes for a window. The four know about each other. When an alert turns a flag off, its ramp pauses at the step that caused it and keeps the rest of its hold, and a freeze holds that ramp while leaving the kill switch armed.
Four controls for a safer release
Each is its own setting with its own guide. Turn on the ones your release process needs.
Team and up
Ramp a change on a timer
A rollout ramp walks a percentage rollout up a ladder you set: 1% for an hour, 5% for four, then 25%, 50% and 100%. Featureflip applies each step when its hold runs out, so nobody has to come back at 2am to move a slider. The ramp stops itself the moment a person edits the rule or the flag goes off, and it keeps the rest of its hold for when you resume.
Team and up
Let monitoring switch it off
A flag trigger is a secret URL that turns one flag off (or on) in one environment. Point a Grafana or Datadog alert at it and the kill switch flips when error rates climb, with no one paged awake to do it. Flag changes flow back the other way too. Slack, Datadog and Grafana are built-in webhook destinations, so the flip shows up on the dashboard your on-call engineer is already watching.
Key check on every plan, reasons on Business
Put a guard on production
A critical environment adds friction where you want it. The dashboard can make you type the environment key before a change goes through, which is enough to catch the toggle somebody flipped in the wrong browser tab. On Business, every change can also require a written reason. The server enforces that one, so it holds for the API, the CLI and AI agents as well as the dashboard, and the reason lands in the audit log next to the change.
Business and up
Freeze a window
A change freeze blocks flag changes in the environments you pick until a set time, and you can schedule one ahead of a sales weekend or hit Freeze now the minute an incident starts. SDKs keep serving flags the whole time. An Owner or Admin can still break the freeze for one change, with a reason, and the audit log records who did it and why.
One release, start to finish
A week in the life of one checkout flag, in a production environment that requires change reasons.
-
Tuesday, 10:00
The ramp starts
You start a ramp on the new checkout flag in production: 1%, 5%, 25%, 50%, 100%, with a few hours between steps. Production requires a change reason, so the ramp asks for one, and "Launch checkout v2, PLAT-881" goes into the audit log with it.
-
Tuesday, 23:40
An alert turns it off
At 25%, checkout errors spike. The Datadog monitor calls a flag trigger, the flag goes off for everyone within seconds, and the ramp pauses at the step that caused it. The flip lands in your Slack channel and on the Datadog timeline next to the alert.
-
Wednesday, 09:15
A person turns it back on
Someone reads the alert, ships the fix, and turns the flag back on with a reason of their own. They resume the ramp. A hold never counts down while its flag is off, so the ramp carries on from 25% with the same time left on the clock as when the alert fired.
-
Friday, 18:00
The sales-weekend freeze begins
A scheduled freeze covers production until Monday. The ramp waits out the freeze and picks up again by itself when it ends. A promo flag that has to go live at midnight Saturday was pre-approved through this window by an Admin, so it applies on time. The trigger stays armed all weekend.
Every step above is a normal flag change. It streams to connected SDKs within seconds, sends a webhook, and shows up on the flag's History tab with who made it and why. The progressive rollouts and kill switch use cases cover each pattern on its own.
Which plan includes what
Ramps and triggers start on Team at $39 a month billed annually. Business, at $199, adds required change reasons and change freezes. Enterprise includes everything on Business.
| Control | Free | Team | Pro | Business |
|---|---|---|---|---|
| Percentage rollouts and kill switches | Yes | Yes | Yes | Yes |
| Type the environment key before a change | Yes | Yes | Yes | Yes |
| Slack, Datadog and Grafana destinations | 1 endpoint | 3 endpoints | 3 endpoints | Unlimited |
| Scheduled changes | No | Yes | Yes | Yes |
| Rollout ramps | No | Yes | Yes | Yes |
| Flag triggers | No | Yes | Yes | Yes |
| Required change reasons | No | No | No | Yes |
| Change freezes | No | No | No | Yes |
| Audit log retention | 7 days | 30 days | 30 days | 90 days |
Nothing is metered. Plans and limits lists what happens to each control if a plan changes. For how other vendors handle the same jobs, the comparison pages check each one against its own docs.
Where each control stops
Plan around these before you rely on the controls in an incident.
- Ramps run on a clock. A ramp holds each step for exactly as long as you told it to, whatever your dashboards say, and only a trigger wired to an alert makes a bad step shut itself off. A metric-gated canary at the deployment layer can catch a regression too small to cross an alert threshold, and the ramps guide on our blog covers when you'd want one.
- A freeze covers one environment's settings. Segment edits and flag-wide edits such as variations reach every environment at once, so a freeze leaves them alone. Aborting a ramp is allowed too, since it can only lower the rollout.
- Triggers run through the Management API. Your SDKs keep evaluating flags if it's unreachable, but an alert can't flip a flag until it's back. Treat each trigger URL like an API key, and rotate it if it leaks.
- Turning a flag back on stays with a person. Turn-on triggers exist, but the automated kill switch post explains why we'd keep that step human.
Frequently asked questions
- What is release safety for feature flags?
- Release safety is the set of controls that limit how much damage a flag change can do and how fast you can undo it. In Featureflip that means four things working together. Rollout ramps raise a percentage rollout on a timer. Flag triggers let a monitoring alert turn a flag off. Critical environments require a reason, or the environment key, before a production change. Change freezes block changes for a window such as a sales weekend. Each one is a separate setting you can use alone. Combined, an alert that turns a flag off also pauses the ramp on it, and a freeze holds that ramp without disarming the kill switch.
- Can a rollout stop itself when an error alert fires?
- Yes, if the alert can reach the flag. Point a Datadog or Grafana alert at a flag trigger that turns the flag off. When the alert fires, the flag goes off for everyone, and any rollout ramp on it pauses with the rest of its hold intact. A person then reads the alert, fixes the cause, turns the flag back on and resumes the ramp. Featureflip ramps run on time. A metric-gated canary can catch a regression too small to cross an alert threshold, and a ramp with a trigger can't.
- Does a change freeze stop my app from getting flag values?
- No. A freeze blocks writes to the environments it covers: toggles, targeting rules, prerequisites, rollout changes, copying configuration in, and starting or advancing a ramp. Reading flags, evaluating them and SDK streaming carry on as normal, so your applications keep getting values the whole time. A blocked change gets an error that names the freeze and when it ends. Segment edits and flag-wide edits such as variations aren't scoped to one environment, so a freeze doesn't block them.
- Does a kill switch still work during a change freeze?
- Yes. Flag triggers fire during a freeze on purpose, because a kill switch that a freeze could block would be no use during an incident. The change goes through and the audit log records it as a pass-through for that window. People can still act too. An Owner or Admin can break the freeze for a single change by giving a reason, and that override is audited with their name and the window it broke.
- Which plan do I need for these features?
- Rollout ramps, flag triggers and scheduled changes are on Team, Pro, Business and Enterprise. Team is $49 a month, or $39 billed annually. Required change reasons and change freezes are on Business and Enterprise, and Business is $249 a month, or $199 billed annually. Typing the environment key before a change, percentage rollouts and the Slack, Datadog and Grafana webhook destinations are on every plan, Free included. Nothing is metered: no seats, no monthly active users and no evaluation caps.
- Can I drive these from the API, the CLI or an AI agent?
- Yes. Ramps, triggers and freeze windows all have endpoints on the public Management API, and the MCP server wraps them as tools (manage_rollout_ramp, manage_flag_trigger, manage_change_freeze), so you can ask an agent to ramp a rule over a week or freeze production until Monday. Required change reasons are enforced by the server, so a script, the CLI or an agent has to send one, through the X-Intent-Reason header or the tool's reason argument. A missing reason gets a clear error asking for one.
Ship the next release with a safety net
Start on Free with percentage rollouts and kill switches. Ramps and triggers come with Team, and change reasons and freezes with Business at $199 a month billed annually.
Related
Progressive rollouts (use case)
The step ladder, and what to watch at each step before you climb.
Kill switch (use case)
Turn a feature off in seconds, by hand or from an alert.
Environments (feature)
Isolated production config, critical environments and freezes in context.
Feature flag governance (blog)
The questions an auditor asks, mapped to the controls that answer them.