Launching today
Astra
Make AI agents that never see your data
87 followers
Make AI agents that never see your data
87 followers
Your AI agent shouldn't see raw sensitive data to do its job. Most of the time it doesn't need to. Astra tokenizes PHI, PCI, and PII before it reaches your agent. It reasons on safe tokens, acts on real values at execution the raw data never touches the model context. Two lines of code. Works with any agent framework.
Astra
Product Hunt
@obed_mpaka1 the tokenization-before-prompt approach is interesting — what happens when the agent's reasoning output references a token and you need to log or audit that decision? Does the audit trail show the real value or does it stay tokenized end-to-end?
Astra
@jimmypk so the audit log stores tokens, not real values.
[CVT:NAME:A1B2] filled first_name at hospital.com at 14:13:22. Authorized. Uses remaining: 0.
The real value lives in one place, the vault. The reveal log records that a reveal happened, not what was revealed. Those two things are deliberately separate. If they needs to know which patient was affected, they run the token through the executor with proper authorization. The audit trail points to the token. The vault holds the value. Nobody hands them a document full of PHI.
Agent reasoning log : tokens only
Audit trail : tokens, action, timestamp, who triggered the reveal
Vault : real values, access-controlled separately
Reveal log : proof a reveal happened, without storing what was revealed
You can hand that audit log to a regulator as-is. It doesn't become a PHI liability the moment you open it.
That's the point.