Release Watch · Team and Business

Ship the flag.
We'll watch the fallout.

Release Watch tells you what to monitor before every flag change, watches it while the change rolls out, and pauses the rollout before your customers notice anything.

new-checkout · Production · 0% → 20%Watching · 3h 12m

Error rate

0.41% vs 0.38% before

OK

p95 latency

+11% of +15% limit

Near

Users getting true

20.4% · 8,310 users

OK

New errors in Sentry

0 since the change

OK

Auto-pause armed. If a signal stays over its limit, the flag goes back to how it was and #checkout-oncall hears why.

How it works

Every flag change is a deploy.
Now someone's watching it.

Before: a watch plan

When someone proposes a change, from the dashboard, the API or an AI agent, Railflag drafts what to monitor: the signals, their limits, how long, and why each one matters for this change.

During: watched every minute

Railflag reads your metrics and compares them with the same window before the change, and counts how many users actually got each variation, so a regression shows up while it affects 20% of users, not 100%.

When it breaks: paused

A heads-up when a signal gets close, an alert when it breaks, and, if you turned it on, an automatic pause that puts the flag back to how it was before the change.

Built to be trusted

Nothing bypasses the rules
your team already trusts.

A pause is an ordinary change through Railflag's API: validated, published to every SDK within a second, and in the audit log with the actor "Release Watch".

Never overwrites a person

If someone changed the flag after the watch started, Release Watch alerts instead of pausing.

Unknown never pauses

If a monitoring tool can't be read, the signal shows as unknown and the rollout is left alone.

Read-only connections

Credentials are entered on a one-time page, encrypted, and used only to read. We never call a tool's write endpoints.

Suggestions without AI

Plans come from fixed rules over flag data and metric names. No user data and no third-party AI service.

Counts, not users

SDKs send how many evaluations each variation got per minute. No user ids. One option turns it off.

Works with your agent

The MCP server suggests a watch plan before a Production change and attaches it, so changes Claude proposes are watched too.

Connections

Reads the tools you already pay for.

Prometheus and Grafana

Any PromQL query, including Grafana Cloud's Prometheus endpoint.

Datadog

Metric queries for error rates, latency and business numbers.

New Relic

NRQL queries through NerdGraph, read-only.

Sentry

New issues since the change, scoped to the countries or plans the change targets.

Slack and email

Heads-ups, alerts and pause notices, once per signal, to the people and channels you choose.

Business

"What should we monitor?"
Answered every Monday.

The coverage report lists live Production flags nobody is watching, ranked by risk, kill switches with no check that they work, flags no SDK has evaluated in 30 days, and metrics your services are missing. It arrives by email, and your agent can ask for it any time.

Plans

On Team and Business.

StarterTeamBusiness
Watch plans, live watching, alerts and auto-pauseLocked previewIncludedIncluded
Monitoring connectionsNone2Unlimited
Watch historyNone30 days2 years, exportable
Weekly coverage reportNoneNoneIncluded
Require a watch for every change in an environmentNoneNoneIncluded

Approvals, the audit log and the MCP server stay on every plan. See pricing.

Frequently asked questions

How does Release Watch decide what to monitor?
From fixed rules applied to what Railflag already knows: the flag's kind and tags, the environment, how the change moves users, and the metrics your connected tools report. Every suggested signal says why it was suggested. No AI service and no user data is involved.
What happens when a limit breaks?
If auto-pause is on, Release Watch puts the flag back to how it was before the change, through Railflag's normal API, so the change is validated, published to every SDK and recorded in the audit log with the actor Release Watch. If the flag was changed again in the meantime, it alerts instead of overwriting that change.
Which monitoring tools does Release Watch support?
Prometheus and Grafana, Datadog, Sentry and New Relic for signals, and Slack and email for alerts. Connections are read-only: Release Watch never calls a tool's write endpoints.
Which plans include Release Watch?
Team includes watch plans, live watching, alerts and auto-pause with 2 monitoring connections. Business adds unlimited connections, the weekly coverage report, two years of watch history and the option to require a watch for every Production change. Starter shows a locked preview.

Put your next rollout under watch.

Start free, then turn on Release Watch with the Team plan when your first Production change needs it.