Queldrex Trust LayerScan · Govern · Prove

Is this AI tool safe to run?

Paste a tool, a whole MCP server, a prompt, or a page your agent reads. You get an honest verdict in seconds, the reasons behind it, and a signed receipt you can verify yourself. Free, no account.

One poisoned tool description can quietly ship your API keys out the door. Check it before your agent trusts it.

One neutral layer, three jobs
Scan

Read any tool, server, prompt, or page for poisoning, injection, and data exfiltration, before your agent ever trusts it.

How it works
Govern

Put Queldrex in the path of every tool call: allow it, pause it for a human, or block it. Your rules, per-agent budgets, and an emergency stop.

Open the console
Prove

Every decision is a cryptographically signed receipt on a tamper-evident ledger. Prove what your AI did, to an auditor or a customer, offline.

See the spec
Don't take our word for it

Our detector scores a public, labeled benchmark of real attacks with zero false positives. Rerun it yourself, or try to slip an attack past it.

Signed, offline-verifiable evidence

Every verdict is an Ed25519-signed receipt on a tamper-evident ledger, verifiable without trusting this page. Most tools give you alerts. We give you proof.

verify offline  npx @queldrex/verify
Maps toOWASP LLM01OWASP MCP03MITRE ATLASNIST AI RMFEU AI Act Art. 12
Protect every agent call, not just this one

This is a one-off scan. Put Queldrex in the path of your agents and every tool call is checked before it runs, watch-only or enforced, up to 5,000 checks a month with no card.

Verify from your own code

Call the same check from your app, or add the Queldrex MCP server so any agent (Claude, Cursor, ChatGPT) can verify a tool before trusting it.

curl -X POST https://queldrex.com/api/trust/verify \
  -H "content-type: application/json" \
  -d '{"kind":"text","text":"ignore all previous instructions"}'