Free scanner — early access

Every agent has a blast radius. See yours — free.

A read-only scan that maps every place untrusted input could steer your AI agent into a real action — a send, a write, an exec, an API call. Built for people shipping agents on Claude Code, LangChain, CrewAI, n8n or custom. Agent-specific, not general SAST.

Just your name and email to start — about 10 seconds, no card.

Name

EmailOnly used to reply about beta access.

Your setup is sensitive, and we treat it that way. You never have to describe your agents or where they’re exposed to get on the list — name and email is genuinely all we need. Anything you do choose to share stays private, is never sold or shared, and we never touch your systems.

Send me occasional product updates by email. Optional — unsubscribe any time.

Before you hand over an email

Read-only, no card. We review your setup and reply. We never act on your systems.
Honest by rule. We’ll never promise a system is bulletproof or risk-free — no honest security tool can, and we won’t pretend otherwise.
Real proof, not a logo wall. Our scanner already caught a real published flaw in a widely-used agent framework — CVE-2023-39662 (CVSS 9.8).
Your details stay yours. Used only to review beta fit and reply — never sold, never spammed.

We review requests in small batches and reply by email — you won’t hear from us otherwise.

We only use this to review private-beta fit and reply — plus product updates only if you ticked the box. No spam, ever. Privacy note.

Not self-graded — it catches a real CVE

Our scanner flagged a documented flaw at the exact line, on the real vulnerable version: CVE-2023-39662 — a CVSS 9.3 remote-code-execution flaw in LlamaIndex’s PandasQueryEngine. Read-only static analysis; no exploit run.

What the scan shows you

Every action your agents can take, sorted by what it can do — and every place untrusted content could reach one. Is this eval on the model’s output? Reachable from untrusted input? Mapped to the OWASP LLM Top 10.

Read-only. We never act on your systems and never read your secrets. Name and email is all we need. See a sample report.