Veridex puts a rule in front of the data your AI agents and apps move: sensitive values are masked, redacted, encrypted or blocked, and every run leaves an Ed25519-signed proof anyone can verify with public keys.
Hi Product Hunt.
I'm Alfa, founder of AlfaNest Labs, a small data security company in France.
Why I built Veridex. AI agents now read files, write summaries, store memory and open tickets. When someone asks later - what did the agent see, what was allowed, and can you prove the record wasn't edited?
A normal log only helps if you trust whoever holds it.
I wanted evidence anyone can check.
What it does:
- Rule before the work runs. Your agent or app sends a JSON payload (or a CSV/text file) to a run. Veridex detects secrets, tokens, IBANs, card numbers, national IDs, contact details and more, then applies MASK, REDACT, ENCRYPT or DENY before the data moves on. Your own - policy Rules - can replace the automatic rules.
- Tool-call gate. Allow, deny, or hold for a human before an agent's tool runs; approvals are resolved in the dashboard, never by the agent's own key.
- Agent identity. Each agent is bound to a runtime hash; a run from a different build is refused until you rotate it.
- Signed proof per run. A JSON proof bundle and a PDF report, signed with Ed25519 (RFC 8032) over canonical JSON (RFC 8785 rules), chained with SHA-256 and a Merkle receipt.
- Anyone can verify. Public keys are published (retired keys stay valid after rotation). Verify a file on the public page, through the API, or offline with an open-source, MIT-licensed Python verifier that checks whole folders at once.
- Auditor access. Give an auditor a read-only token, limited in time and scope; you see every request they make.
Framework mapping. Each proof carries results for 17 controls mapped to GDPR, ISO 27001, SOC 2, HIPAA, MiCA, ISO 42001 and NIST AI RMF, so you see what is covered and what stays with you.
Try it: sign in and the demo tier is there (500 runs, 5 agents). The docs, the signature format and the test vectors are all public at alfanestlabs.com/veridex/docs, so you can write your own verifier if you prefer.
I'd really like feedback from people running agents in production: what would you want in a receipt that isn't there yet? I'll be here all day answering questions.