"Every 14 minutes, a developer leaks AI intellectual property to public GitHub. Most tools scan after push. Provn stops that leak before the commit lands, fully locally."
Open terminal
"Here’s a staged change with a hidden system prompt and a credential."
Run: git commit -m "add feature"
Terminal shows a block with file, line, tier, and redacted preview
"Provn caught it before the code left the laptop."
Show auto-redact preview
"Provn doesn’t just say no. It shows exactly what is dangerous and offers a safe replacement."
Accept the redaction or fix the file and re-run the commit
"Now the commit passes."
Show benchmark output or report
"We measure this against curated leak cases, not hand-wavy examples."
- strong recall on seeded leaks
- low false positive rate
- fast clean-commit latency
"And the semantic layer is optional, local, and fail-safe."
"Every decision is written to a tamper-evident audit chain. If someone bypasses the hook, we can still verify what happened later."
Run: provn verify-audit
"That gives teams both prevention and accountability."
"Provn is a local-first AI security guard for your repo: fast, auditable, and built for the era of prompt leaks and secret spills."
"Provn is designed to make bypasses visible through audit logging and CI re-checks. The local hook is the first line of defense, not the only one."
"We use layered detection, configurable policy tiers, and LeakBench-style evaluation so we can tune against real examples instead of guessing."
"If you are protecting sensitive code, sending it to a cloud scanner is a contradiction. Provn keeps scanning on the machine by default."
"GitHub scans after push. Provn scans before commit. That changes the blast radius completely."
"It gives us a local semantic layer that can run on developer hardware without requiring a hosted inference API."
"Because the CLI is the real product wedge. It is the fastest path to protection, lowest maintenance surface, and easiest thing for a team to adopt."