Define
Model flags and targeting rules with typed primitives in your repository.
Release control as a function
Decide with
explicit logic.
Define feature access as code. shipd evaluates users, environments, and rollout rules without hiding the decision.
decisionf(u, e, t) → {0, 1}
user × environment × time
f(mira, production, now)= 1Local evaluation
0 network hops
Deterministic rules
Full decision trace
No black box between
the rule and the result.
Change the inputs below. The result updates from the same conditions your application evaluates.
Live evaluationnew-checkout
request_8f31c01
country∈{ US, CA, GB }true02
plan=protrue03
bucket(user)<rollouttrueThe logic starts
in your codebase.
Types, defaults, variants, ownership, and lifecycle stay reviewable beside the software they control.
01 export const flags = defineFlags({
02 newCheckout: flag.boolean({
03 default: false,
04 rules: [{
05 when: country.in("US", "CA", "GB"),
06 and: plan.eq("pro"),
07 rollout: 42.7,
08 serve: true,
09 }],
10 }),
11 });Definition valid
1 rule · 2 variantsFrom definition
to decision.
Review
See the exact environment plan in CI before a definition reaches production.
Publish
Ship a signed, versioned snapshot to every connected runtime.
Evaluate
Resolve each request locally from explicit inputs and deterministic rules.
04 / Environment plan
Know what changes before it changes.
Every pull request produces a readable plan across environments. Review the audience, rollout, and resulting definition together.
Feature flag best practicesplan_7a3fenvironmentcurrentproposedstatus
development100%100%unchanged
previewinternalinternal + betachange
production31.4%42.7%change
Fast enough to sit
inside every request.
The SDK resolves the snapshot locally. No extra hop on the request.
eval_f412Snapshot loaded
Context normalized
Rule 01 matched
Bucket 0.314 < 0.427
Engineering teams that want flags in git.
shipd is for teams who keep flag keys typed beside the code they gate, review environment plans in pull requests, and evaluate each request locally—not for teams that need a dashboard-first platform.
The next release can be explicit.