Ship AI-written code
without shipping a security leak.
AI writes most of your PRs now — and quietly slips in SQL injection, hardcoded secrets, SSRF and broken crypto. slopGrade Firewall catches high-severity leaks in CI, before merge — deterministic, $0, and only a {file · line · class} fingerprint is sent, never the contents of your files.
Built under the constraints of regulated teams — fintech · health · defense · public sector.
The security scan that never sees your source code.
The SaaS scanners that catch these leaks ingest your source code into their platform. For a regulated team — fintech, health, defense — that alone is a dealbreaker. slopGrade Firewall runs the detectors inside your runner and posts only a structural fingerprint for classification.
Typical SaaS scanner
Your whole repo — every line, every secret — copied into a third-party platform.
slopGrade Firewall
A structural fingerprint — path, line, class. The file content never leaves. Audit it with --print-payload.
Detect → adapt → verify. Nothing else leaves.
Locally, in your runner
The open-source client walks the diff and builds a structural fingerprint — file paths + {file, line, kind}. The contents of your files never leave the machine.
Server verdict + gate
The fingerprint is classified server-side. A hard leak blocks the check on a paid private repo; public repos are gated free. The paywall fails open — never blocked by mistake.
Fix computed in your runner
Every finding posts inline on its line. The auto-fix is generated AND verified in your runner — the server never touches it, and it's only offered once it's proven to clear the leak.
The 11 highest-severity classes — free.
Across JavaScript/TypeScript, Python, Go and .NET. The hosted product runs the full 124-class catalog with cross-file dataflow.
The hosted product adds +113 classes and cross-file dataflow →
Powered by the Heisen engine
The hosted gate runs on a deterministic dataflow engine that proves each finding across 5 languages — self-hostable and air-gapped, $0 per scan. It doesn't pattern-match; it follows the data.
The free tier is real. The gate is where it pays.
Block leaks on private repos. Priced per repo, never per seat.
Users unlimited. The per-repo price drops with volume. 14-day free trial — no card.
Public repos are gated free, forever. Private-repo blocking starts at $11/repo/mo · 14-day free trial, no card.
Ready to block the next leak before it merges?
Add it to one workflow file. The first AI-written PR you scan will tell you what it was about to let through.
The questions a security team asks first.
Does my source code leave the runner?
The contents of your files never leave the runner. The client reads files locally and posts only a structural fingerprint — paths + {file, line, kind}. Run node isolation-gate.mjs --print-payload to audit the exact bytes sent. Even a hardcoded secret emits only the location and kind, never the value.
Will it spam my PRs or break my build?
No. Start in advisory mode — you see every leak without blocking anything — then switch to gate when you're confident. The gate fails open by default: a network, server or OIDC failure never breaks a build. Only a real, reliable hard leak on a paid private repo blocks; strict lets you fail closed.
The auto-fix — where does it run?
In your runner. The fix is generated and verified locally (we re-run the detector on the patched buffer and keep the fix only if it clears the leak) — the server never touches it, and your tree is never modified: applying it stays your call.
Can I audit the code that runs in my CI?
Yes — the client is open source (MIT). Pin the commit SHA to freeze the exact code (a tag can be re-pointed, a SHA can't). Zero npm dependencies; only Node builtins.
What languages?
JavaScript/TypeScript, Python, Go and .NET today, across the 11 open-source classes; the paid firewall adds the full Go/.NET/Java breadth.