[ Infrastructure for AI software reliability ]
Your AI writes the code.
Grace keeps control of it.
Grace sits between your AI agents and your production app. It holds every change to your rules, stops the risky ones before they land, and only puts its seal on what it can prove.
Free on one app · No credit card · Code hosted in France
WHAT WILL BREAK FIRST
- CRITICALStripe webhook accepts unsigned eventsapp/api/stripe/route.ts:42
- CRITICALService key shipped in client bundlelib/supabase.ts:7
- HIGHNo rate limit on the signup endpointapp/api/auth/signup/route.ts
- HIGHCheckout has no test, 14 commits last monthcomponents/Checkout.tsx
- MEDIUMThree copies of the same date helperlib/date.ts · utils/format.ts
Anyone can POST a fake “payment succeeded” to this route today. I check the signature with your existing secret and reject the rest. Nothing else changes.
GRACE GUARDS THE AGENTS AND THE STACK YOU ALREADY SHIP WITH
- GitHub
- Claude Code
- Cursor
- Codex
- Sentry
- Vercel
- Supabase
- Lovable
- Bolt
[ Meet the crew ]
Grace leads.
The specialists are called in.
Each change goes to the agent who knows that part of your app. Grace coordinates them, remembers what was decided, and is the only one who signs off. You always see who is working, and why.
- ProductFrames the change
- Lead devDecides how to build it
- SecurityAuth, secrets, access
- GraceLeads, remembers, signs off
- DataModels and migrations
- QATests before and after
- AuditorRanks what breaks first
Know exactly what you're shipping.
Grace reads the repository itself: the source, the history, who touched what. It scores your app's health and ranks what will break first, in plain words, before your users find it.
- Reads code, history and ownership, not HTTP responses
- Ranked by what breaks first, not by alert count
- A full health check every month, free on one app
No proof, no change.
Every fix comes with a test that fails before and passes after. Run it in Grace, or let your own AI do the work through the Grace MCP: Grace still checks the result before anything is committed.
- One pull request per fix, nothing merged without you
- Tests run before and after, impact checked
- Works with Claude Code, Cursor and Codex via MCP
> fix the critical findings Grace left on this repo
● grace_fix_brief (grace MCP)
2 groups by cause · wave 1 can run in parallel
🧠 Grace — lead dev: start with the webhook. Same cause as the signup route: input trusted before it is checked.
● Edit app/api/stripe/route.ts (+4 −1)
● grace_review_submit
🧠 Grace — guardian: tests 48/48 before and after. The secret is read from env, not inlined. Safe to commit.
✓ Opened PR #214 fix(stripe): verify webhook signatures
Old problems stay fixed. New ones never land.
Grace runs inside Claude Code and Cursor and checks your agents' edits before they are committed. Every push is held to your project's rules and recorded decisions, and anything that brings an old bug back is blocked.
- A check on every push and pull request
- Your AI's edits checked before commit
- Included in every plan, Free too
- buildPassed in 1m 12s
- tests212 passed
- grace / guard1 regression blocked
- grace / memoryMatches 3 team decisions
This brings back the bug fixed in #214 in a new place: the invite token is read from the URL and trusted before it is checked. Same fix, one line, I can push it to this branch.
[ What earns the seal ]
Nothing gets the seal by default.
Before Grace signs off on a change, every specialist it touches has to agree. One check fails, no seal, and Grace tells you why in plain words.
- No secret in code, no route left openAuth checked on every endpoint the change touches✓ checked
- Tests pass before and afterAnd a test is added when there was none✓ checked
- It fits how your code is builtConventions, patterns, the decisions your team recorded✓ checked
- Migrations can be undoneNo column dropped that something still reads✓ checked
- No old bug comes backEvery past fix is checked again✓ checked
- You stay the one who mergesThe seal is a recommendation. Your click is the decision✓ checked
[ Why Grace ]
They scan your URL.
Grace reads your repository.
An outside scanner sees what a request returns. It cannot see an unused code path, a secret in a file that never reaches the browser, the history of a module, or why a trade-off was made.
| LEAVE IT FRAGILE | URL SCANNERS & GENERIC TOOLS | GRACE | |
|---|---|---|---|
| What it reads | Nothing | Your live URL, from outside | Your repository: code, history, owners |
| What breaks first | You find out in production | A long list of unranked alerts | Ranked before it breaks |
| The fixes | Nobody dares touch it | Raw suggestions you integrate | Tested, reversible pull requests |
| Control | Black box | Auto-merge or nothing | You approve every change |
| Next month | Same mess again | Same scan again | Guarded: fixed stays fixed |
Grace remembers why.
Every decision and fix is kept, so the next person, or the next AI, doesn't start from zero.
Grace doesn't replace the AI that writes your code. It keeps that AI under control.
[ From the people who run it ]
Why Grace exists.
Picking up code you didn't write is the hard part of this job, and it almost never arrives clean. Grace is that experience turned into a product: a guardian angel for the app you already have.
[ Pricing ]
Free on one app. $29 per app when you carry more.
Free is the real product, not a teaser. Health checks never cost credits. You pay for what accumulates: more apps, unlimited health checks, and a memory that never resets.
Free
$0
no card, no commitment
- 1 app, 1 full health check a month
- Every finding shown
- Grace in your editor: Claude Code or Cursor
- 5 fix credits a month
- Memory of your last 30 days
Pro
$29
per app / month · 20 credits per app
- Unlimited health checks, run automatically every week
- Memory that never resets
- Portfolio view over every app
- Reliability and ROI proofs
- White-label report
Agency
$199
per month · 5 apps, then $15 per app
- 60 pooled credits, +10 per extra app
- Daily health check per app
- Grace in your editor, on every app
- One view over every client
- Priority support
Enterprise
Custom
past twenty apps
- Credit allowance sized to your portfolio
- Number of apps set by contract
- Daily health check + report per app
- Grace in your editor + white-label on every app
- Dedicated support and onboarding
A fix is one credit, a heavy one two or three, and the estimate is shown before anything runs. Open the price simulator →
Yes. Connect your repo and Grace produces a health report and a ranked risk list in minutes. No card, no commitment.
[ Your turn ]
Get your first
Grace Approved.
Connect a repo. In minutes, the crew tells you what will break first.
Free on one app, no credit card.
