// shift-left + shift-right · QA on both ends

Avoid Jank at Build time, find Jank at Runtime

Jank moves quality checks so far left they happen before the developer even has to think about them — and so far right they catch the bugs that escape staging, CI, and human QA. Same audit on both ends. Zero work either way.

Complements your existing tests, lint, CI, and human QA — never replaces them.

Where Jank sits in your pipeline

PLUGIN

free · local

→

AI codes

→

tests + lint

→

CI / review

→

QA / staging

→

production

→

CLOUD

hosted

← shift-left shift-right →

Plugin runs inside the AI coding session — every keystroke gets a quality check before the developer sees the code. Cloud runs against staging or production — catches what slipped through everything in the middle plus the full-stack issues no static analysis can ever find.

⚡ Shift-left · free

Jank Plugin

Catch issues before they leave the developer's machine. The plugin runs inside Claude Code, Cursor, Codex — auditing every change the AI makes against 7 quality categories (security, perf, a11y, reliability, content, privacy, code quality) using both static analysis and real Chrome with 6 user personas.

Most efficient time to fix a bug is before it's committed. The plugin makes that the default — no extra step in the workflow.

/jank — full sweep · /jank_light — instant static · /jank_test <feat> — surgical · /jank_explore — 5 user agents. See plugin ROI →

📊 Shift-right · hosted

Jank Cloud

Catch what escapes everything else. Cloud runs the same audit against your live staging or production URL — finds the bugs that only exist when the full stack is actually executing: third-party scripts, real network conditions, real cookies, real auth, real CDN behavior. None of these show up in code review or unit tests.

It's the layer below your existing E2E and synthetic-monitoring stack — auditing what shipped, not what was supposed to ship.

AI vision-LLM audit · 6 personas · live test flows · subpages · WCAG · email + Slack alerts. See cloud ROI →

Run a Cloud Report free · trial run

Enter a public URL plus your email — we'll create your free account (1 free trial report) and queue this report automatically. First-time visitors get bounced to a 30-second signup; returning users with an API key submit immediately.

URL to test Your email

Advanced options (provider, personas, tunnel, label, …)

Visibility

Label (optional)

LLM Provider

LLM Model

Also email these recipients (comma-separated)

Tunnel (optional)

Most sites don't need this — leave it on "none." Jank runs in the cloud, so it can only reach addresses that exist on the public internet. If your site lives on localhost, a dev laptop, a corporate VPN, or an internal staging server, pick a tunnel below. A tunnel briefly forwards traffic from a public hostname to your machine for the run, then disconnects. Which one should I pick? →

Behind a firewall? Pick one:

Application-layer tunnels (HTTP/S only)

Easiest. Just exposes a single web app.

Cloudflare Tunnel — free · zero-trust · cloudflared tunnel --url http://localhost:8080 · docs
ngrok — ngrok http 8080 · docs
Tailscale Funnel — auth keys + tailscale funnel 8080 · docs
SSH reverse tunnel — ssh -R 8080:localhost:80 user@bastion · docs

Network-layer VPNs (full network access)

For internal IPs and multi-service apps. Requires a self-hosted runner — Cloud Run can't run network VPNs without elevated caps.

WireGuard — modern, fast UDP VPN. Add Jank's runner as a peer. docs
OpenVPN — paste a .ovpn profile. docker run -v ./vpn:/vpn --cap-add=NET_ADMIN ...
IPsec / IKEv2 — Cisco / strongSwan / pfSense gateways. Strict cert or PSK auth.

GCP-native

VPC Serverless Connector — direct route to your GCP VPC for Cloud Run. gcloud run deploy --vpc-connector=NAME --vpc-egress=all-traffic · docs

Self-managed (you start the tunnel)

localtunnel — npx localtunnel --port 8080
bore — bore local 8080 --to bore.pub (Rust, no signup)
localhost.run — ssh -R 80:localhost:8080 nokey@localhost.run (no install)
serveo — ssh -R 80:localhost:8080 serveo.net (no install)
pinggy — ssh -p 443 -R 0:localhost:8080 a.pinggy.io
Microsoft Dev Tunnels — devtunnel host -p 8080 --allow-anonymous
VS Code port forwarding — Ports panel → Forward → "Make Public"
GitHub Codespaces — Ports tab → right-click → Visibility: Public
inlets — self-hosted on a $5 VPS
frp — fast reverse proxy, self-hosted

URL rewriting: If your test target URL is http://localhost:3000 or http://internal-app.company.com, paste the public tunnel/VPC URL into the "Tunnel URL" field. The runner rewrites at execution time so AI evaluation hits the right host.

Self-hosted Docker example: docker run -p 8080:8080 --net=host -e OPENAI_API_KEY=sk-... jank/runner

Test flows

AI proposes realistic user flows on the page (click around, fill forms, navigate) and the runner replays them in a fresh browser. Only flows whose every step passes are shown in the report.

Generate & replay test flows (recommended)

How many flows?

Custom flow context (optional)

Persona feedback

Generate diverse user personas and collect their reactions to the page.

Enable persona feedback (slower run, richer insights)

How it works

We launch headless Chrome, navigate to your URL, capture a full-page screenshot, console messages, and network activity, then hand it all to a vision-capable LLM that scores the page from 0–100 and lists every visible bug ranked by interestingness (security/perf high, routine a11y low).

Submit reports from the command line (curl) — for CI / scripting

Replace $KEY with your unlock key (or omit the header for demo mode).

curl -X POST https://reports.jank.ai/api/reports \
  -H 'Content-Type: application/json' \
  -H 'X-Api-Key: $KEY' \
  -H 'X-Account: you@company.com' \
  -d '{
    "urls": ["https://example.com"],
    "visibility": "public",
    "personas": { "enabled": true, "count": 4 }
  }'

Response includes a reports[].id and reportUrl; poll GET /api/reports/<id> until status:"done".

📈 Pricing 📊 Dashboard 🌐 jank.ai