AI QA on a real copy of your stack

Your app runs the gauntlet before your users do.

Gauntlet deploys your app with its database, payments and sign-in into a throwaway environment. An AI agent tests it like a user and files bugs with proof.

Request access Watch a run

Try it on your machine Free, no account

curl -fsSL https://usegauntlet.dev/install.sh | sh
gauntlet demo

Needs Docker with 4 CPUs and 8 GB. x86-64 Linux or WSL2. AI runs use your Anthropic key.

High Demo store, planted bug

3 notebook packs charged $40.69, but the order says $50.69

Checkout Stripe mock: Notebook 3‑pack × 3$35.70; Shipping$4.99; Charged$40.69

Order SF-1004: 3 × $16.90$50.70; FIVEOFF−$5.00; Shipping$4.99; Total$50.69

With video, Stripe call, Mongo order, payments log and repro.spec.ts. See the run

How AI end-to-end testing works

Describe your stack

Your Helm chart, compose file or Dockerfiles, plus kits: Google sign-in, Stripe, email, MongoDB, Postgres, Redis, GCS.
```
app:
  type: helm
  helm: { chart: ./chart }
kits:
  google: { kit: google-oauth }
  db: { kit: mongodb }
  pay: { kit: stripe }
```

Say what to test

Goals in plain words, not selectors.

journeys:
  - name: Checkout with a coupon
    prompt: Pay with a coupon,
      compare charge and order.

Run it, get bugs with proof

From your terminal, CI or Claude Code. A second agent reproduces each serious bug and checks the source.
```
$ gauntlet run --wait
  failed · 2 high · 1 medium
$ gauntlet mcp # Claude Code
```