You ship faster than memory
Teams — and coding agents — now merge more change in a day than any release process can hold in its head.
Agent-native safe rollouts
Your AI release engineer for every feature you ship.
Fireweave wraps every feature behind managed flags, wires adoption, health, and latency guardrails, then ramps exposure from 1% to 100%. If a guardrail breaks, the controller rolls back by flipping the flag off. No redeploy. No mystery. No babysitting.
Fireweave · rollout board live
Feature Exposure Adoption Health Latency Status
checkout-v2-flow prod 48% 0.02% +3ms Ramping
search-rerank prod 91% 0.01% -8ms Complete
billing-proration prod 12% 0.00% +1ms Soaking
onboarding-ai-hints staging 63% 1.9% +41ms Rolled back
controller guardrail breach → halt → flip flag off → exposure 0%
The problem
You ship faster than you can remember.
Production should not be a black box. The invisible backlog of shipped change is the villain — you lose the ability to answer what shipped, who is exposed, is it healthy, and how to undo it.
APM shows services, not features
Dashboards tell you a service is slow. They can't tell you which shipped feature caused it.
Flags show switches, not safety
A flag tells you a toggle is on. It doesn't tell you whether the feature behind it is healthy.
Rollback means a scramble
When an alert fires, reverting a PR is slow and risky. You need the change one action away from off.
Feature visibility
A live production record for every feature you ship.
Flag tools show flags. APM shows servers. Fireweave shows features — what shipped, where it is live, and whether it is safe.
Sort or filter your shipped change
RecentRiskyRampingCompletedRolled back
- Every shipped feature gets a live production record.
- Sort by recent, risky, ramping, completed, or rolled back.
- Any risky feature is one rollback action away from off.
checkout-v2-flow Ramping · 25%
Rollout 25% exposure
Adoption 48% of cohort
Health 0.02% error
Latency +3ms p95
Environment prod · us-east
Owner @checkout-team
Participants 4 repos live
Status Ramping
How it works
From code change to controlled rollout.
Without hand-wiring the release process. Nine moves take a raw change to a guarded, ramping rollout.
- 01
Create or join a rollout
Start a rollout, or attach your change to an existing one for the same feature.
- 02
Scope & wrap
Scope the feature surface and wrap the code with a managed flag at the right wrap point.
- 03
Key the cohort
Choose a stable cohort key so a user stays on the same side of the flag over time.
- 04
Wire guardrails
Attach adoption, health/error, and latency metrics as guardrails for the ramp.
- 05
Verify & seal
Run verification checks — no orphan flags, safe defaults, telemetry complete — then seal the rollout.
- 06
Deploy gate
The gate waits until every required participant is live before exposure starts.
- 07
Ramp & soak
Ramp through 1% → 5% → 25% → 50% → 100%, soaking at each stage while metrics prove health.
- 08
Complete or roll back
Stay healthy and complete, or breach a guardrail and flip the flag off — no redeploy.
The mental model
The rollout is the unit of shipping.
Not the PR. A rollout owns everything a change needs to ship safely, and PRs are just participants inside it.
one feature = one rollout = one coherent flag set
A rollout owns
FlagsWrap pointsMetricsGuardrailsRamp scheduleParticipantsDeploy stateLifecycle
Rollout
A managed transition from old to new behavior that grows in stages, watches guardrails, and ends completed or rolled back.
Feature flag
The switch in code that controls whether a cohort sees the new behavior. Fireweave manages the percentage for you.
Cohort key
A stable identifier that keeps a user, org, or session on the same side of the flag over time.
Wrap point
The exact code location where Fireweave guards behavior — function, route, component, method, or shadow call.
Guardrail
A metric threshold that decides whether the rollout can continue — error rate, p95 latency, conversion, adoption.
Deploy gate
The checkpoint that waits for all required participants to be live before exposure ramps.
Guardrails & telemetry
Every rollout carries the telemetry to decide.
Guardrails are how Fireweave makes safe decisions: continue, pause, or roll back — never a guess.
Adoption
Are users actually reaching the new path?
Health / error
Is the feature hurting reliability?
Latency
Is the new path slower than baseline?
Beyond the canonical three
ConversionBusiness eventsMismatch counters for refactorsCustom reliability thresholds
Verified before a single user is exposed
- Cohort keying
- No orphan flags
- Safe defaults
- No mixed provider calls
- Telemetry completeness
- Config schema
- Provider health
Automatic ramp & rollback
If a guardrail breaks, Fireweave flips the feature off.
Reverting means flipping a flag, not reverting commits. No redeploy — and the audit trail stays intact.
exposure ramp soak between stages
1%
5%
25%
50%
100%
0%
1% → 5% → 25% → 50% → 100% then guardrail breach → halt → flag off → 0%
Ramp styles
Linear percentageCohort-basedExperiment w/ controlShadowDark launch
Agent-native workflow
The rollout starts where the code changes.
Inside your coding agent. Developers should not hand-write rollout YAML, invent flag keys, or wire metric queries by hand.
- The skill runs inside your coding agent — not a separate release console.
- The agent asks confirmation questions before it changes anything.
- It applies diffs for review and runs verification checks.
- The rollout-ready flow front-loads setup so later shipping is “promote, not wrap.”
Works inside
CursorClaude CodeCodexOpenCodeCline
agent · /fireweave:safe-rollout
you wrap the new checkout flow behind a safe rollout
fw Scoping checkout() · cohort key org_id. Confirm? y
+ const on = await fw.flag("checkout-v2", { key: org.id }); + if (on) return checkoutV2(cart); return checkoutV1(cart);
fw Guardrails wired: error · p95 · adoption
✓ cohort keying ✓ safe defaults ✓ no orphan flags ✓ telemetry complete
fw Rollout checkout-v2-flow registered · waiting on deploy gate ▋
Multi-repo & multi-participant
Complex changes stay coherent.
Related changes join the same rollout instead of spawning duplicates. The deploy gate waits for every required participant before the ramp begins.
web-app feat/checkout-v2 live
api-gateway feat/checkout-v2 live
mobile-sdk feat/checkout-v2 deploying
payments/ledger feat/checkout-v2 live
one rollout checkout-v2-flow coherence group · flags move together
Integrations
Connect the providers your rollout needs.
Flags, metrics, logs, traces, alerts, and deploy state — Fireweave orchestrates the tools you already run.
Managed default providers roll out with status badges — we never say “available today” for anything still shipping.
Cleanup
Flags should protect releases, not become archaeology.
- Completed rollouts should not leave permanent flag rot.
- After a clean 100% soak, stable flags become cleanup candidates.
- Cleanup is measured and confirmed — not reckless deletion.
- Safe shipping includes safe retirement.
search-rerank 100% · soaked clean
Ship behind guardrails Ramp with confidence.
Ramp with confidence.
Roll back without redeploying.
Every feature wrapped, measured, ramped, watched, and kept one action away from rollback — from inside the coding agent you already use.