new Guard is now generally available — explore the dashboard →
Guard · surface 3 of 4

Every merge safe.
Every deploy watched.

Guard gates pull requests with affected-scenario selection, monitors production with safe synthetics, and — when something breaks — lets Mender draft the fix. All without touching a real customer.

5 min setup No real users touched SOC 2 Type II Self-host available
guard — watch — acme-app
live
Trusted by engineering teams at RampLinearBrexMercuryPlaidVercel and 4 YC S25 startups
How Guard Works

Three systems. One release loop.

Guard isn't a test runner — it's a release-safety system. It understands your code graph, picks the right scenarios, runs them safely against cloned vendors, and fixes what it breaks.

① Before Merge

PR gating

Affected-scenario selection picks only the tests touched by your diff. Ancestor-green cache skips what's proven safe. Full-suite fallback when the graph isn't confident.

median check: 2m 14s
② After Merge

Production monitoring

Scheduled synthetics across 5+ regions, running real user flows against your live app — with cloned vendors so no real customer is ever touched.

40 checks / min
③ When it breaks

Mender

Triages the failure, classifies it (product bug? vendor drift? flake?), and — for real bugs — drafts a fix PR with a regression test. A human always approves.

68% acceptance · 14m MTTR
Affected-scenario selection

Run 7 scenarios, not 47.

Guard's dependency graph knows which files map to which scenarios. When a PR touches discount.ts, it runs checkout.coupon and checkout.flow — not your entire suite.

STEP 01

Graph analysis

Guard reads your repo's import graph, route map, and scenario definitions to build a dependency matrix. When a commit lands, it traces which scenarios are affected by the changed files.

Affected-graph trace
base SHAgreen@4a9c11
head SHAa3f29c1
files changedsrc/checkout/discount.ts
scenarios matched7 of 47
time saved~6.2 min
STEP 02

Ancestor-green cache

If a scenario already passed on a parent commit and none of its dependencies changed, Guard skips it. Cache policy is per-scenario — content-hash for static tests, ancestor-green for flow tests.

Cache result
auth.logincached · ancestor green
checkout.flowmiss · ran fresh
checkout.couponmiss · ran fresh → FAIL
overall hit rate61%
STEP 03

Full-suite fallback

When the graph confidence drops — a lockfile touches 50 transitive deps, or a new file has no edges yet — Guard falls back to the full suite automatically. No config. No missed regressions.

Fallback trigger
PR #417 · chore/depsfull suite (47)
reasonlockfile → 50+ transitive deps
graph confidencebelow threshold
resultall 47 passed · 3m 58s
Mender · AI remediation

Don't just find the bug. Fix it.

When Guard catches a regression, Mender reads the failing trace, the commit diff, and the git history — then drafts a fix PR with a regression test. A human always approves. Governance controls what files Mender may touch.

src/checkout/discount.ts◆ mender · confidence 94%
@@ -42,7 +42,7 @@export function applyCoupon(amount, coupon) {
if (!coupon || !coupon.value) return amount;
- return amount * coupon.value;
+ return amount * (1 - coupon.value);
}
+ regression test
+test('SAVE10 yields 10% off, not 10x', () => {
+ expect(applyCoupon(1000, { value: 0.10 })).toBe(900);
+});
Mender reasoning
◆ read src/checkout/discount.ts ◆ grep "coupon.value" → 3 matches ◆ reason coupon.value = 0.10 (fraction) but used as multiplier → 10× price ◆ fix 1 line · discount.ts ◆ test sandbox → ✓ total $9.00 ◆ governance forbidden-paths: clear
Governance rails
Forbidden-paths check
Monthly token budget (2.1 / 5M)
Circuit breaker armed
Human approval required
src/payments/** → blocked
Rejection feeds RLHF loop
Production monitoring

Scenarios × regions. Always running.

Guard runs your critical paths on a cadence against live production — with synthetic identities and cloned vendors, so checks never generate a real charge, a real email, or a real analytics event.

Scenarious-eastus-westeu-westeu-centralap-south
Checkout — card → receipt
checkout.flow · p95 1.9s
Auth — login + 2FA
auth.login · p95 0.7s
Billing — stripe webhook → DB
billing.webhook · p95 1.1s
Email — receipt delivery
email.transactional · p95 3.6s
!
Dashboard — first paint
dashboard.load · p95 1.0s
·
·
operational ! degraded failing shadow-diff
The Molar Suite

Four surfaces. One platform.

Guard is surface 3 of Molar's vision: a complete quality fabric for the AI-native engineering team. Each surface works alone — and they compound together.

SURFACE 01

Cartographer

Map your codebase. Understand how modules connect, which paths carry risk, and where agent-written code lands.

SURFACE 02

Clones

Behavioral replicas of your third-party dependencies. Same API, same error modes, zero real effects. The foundation Guard runs on.

SURFACE 03

Guard

Gate every merge. Watch every deploy. Fix what breaks. The release-safety system for AI-native teams.

SURFACE 04

Trace

Follow a request from browser to database and back. Distributed tracing built on the Molar graph. Coming soon.

By the numbers

Real results from real teams.

84%
of PR checks use a subset — no full-suite bottleneck
2:14
median PR check time — down from 38 min full suite
14m
mean time to resolve — 66% faster with Mender
0
real customers touched by synthetic checks — ever
Also known as

PR gating & synthetics — with auto-fix.

Guard targets teams comparing Checkly, Datadog Synthetics, and deployment gates — but adds affected-scenario selection, cloned-vendor safe synthetics, and Mender fix PRs.

PR gating production synthetics deployment gate Checkly alternative CI quality gate auto-fix PR
How is Guard different from Checkly or Datadog Synthetics?

Checkly and Datadog run synthetics against live vendors and real users. Guard runs flows with Clones so production checks never charge a card or email a customer — and adds affected-scenario PR gating plus Trace replay on failure.

What is affected-scenario PR gating?

Guard maps changed files to scenarios via a dependency graph. A PR touching discount.ts runs checkout scenarios — not your full 47-test suite. Median check time: 2m 14s. See GitHub Actions integration.

Does Guard touch real customers during production checks?

No. Production synthetics use cloned Stripe, Twilio, and email vendors. Zero real-user side effects — ever.

What is Mender?

When Guard confirms a regression, Mender triages the failure, classifies it (product bug vs flake vs vendor drift), and drafts a fix PR with a regression test. A human always approves the merge.

Pricing

Pay by the run. Not the runaround.

Every plan includes PR gating, production monitoring, the full dashboard, and the same Mender. The only knobs are runs and seats. Free until you ship — then the math is on this page.

Starter

For solo devs & side projects

$0/ month
  • 500 PR runs / mo
  • 2,000 production runs / mo
  • 2 regions
  • Suggestive Mender (500k tokens)
  • 1 seat
  • Community support
Start for free
Team · most popular

For shipping teams

$299/ month
  • 25,000 PR runs / mo
  • 250,000 production runs / mo
  • 8 regions
  • Aggressive Mender (5M tokens)
  • 15 seats
  • Shadow-prod
  • Priority support · Slack Connect
Start free trial
Business

For orgs that can't break prod

Custom
  • Unlimited runs
  • Unlimited regions
  • SSO / SCIM / RBAC
  • Self-host / in-VPC
  • Data residency
  • Custom Mender budgets
  • Dedicated CSM · SLA
Talk to sales

Stop shipping scared.

Every merge safe. Every deploy watched. Every fix drafted.
Guard is the release-safety system for AI-native teams.

Start free trial See a 90-second tour Read the docs ↗