Security
VerifiedIsolation you can prove, not policy you must trust.
Three independently enforced layers keep every tenant apart. We probe them adversarially and sign the result into a receipt you can reproduce without trusting us.
The receipt
Reproduce it offline, no trust required
The receipt is signed with Ed25519. Verify it against our published public key (a JWK at hunta.ai/.well-known/jwks.json) and re-run the probes against your own tenant.
$ curl -sX POST https://mcp.hunta.ai/v1/admin/verify-isolation -H "Authorization: Bearer <YOUR_KEY>"
fetch receipt ............ ok
verify ed25519 signature . ok
re-run isolation probes .. 0 leaks
compare to attestation ... match
verdict: VERIFIEDBoundary manifest
What is plaintext, and why
Honesty over hand-waving. Some columns stay plaintext by necessity. Here is every one, and the reason.
Severity and timestamps are plaintext on purpose. Ranking facts by severity and scanning event-time ranges both require the column outside the ciphertext. Values stay encrypted, tenant id is bound into the cipher, and the isolation receipt covers all of it.
Isolation you can prove, in five minutes.
Create a key, connect your agent, and reproduce the signed receipt yourself.