Privacy-first AI memory
for fintech & healthcare teams

Whenever a regulated team evaluates an AI memory tool, the same three questions come up — and most tools fail at least one of them:

• Where does the data physically live? US-hosted by default is an immediate blocker under GDPR.
• Can sensitive values (PII, secrets) be stopped before they leave our machines? Server-side redaction isn't enough — the raw value still crossed the wire.
• Can the whole team share knowledge without sharing the underlying sensitive data?

cachly was built so the answer to all three is yes. Let's walk through each.

cachly's EU instances run exclusively on Hetzner's German data centres. There is no "EU option you have to remember to tick" — for EU customers, data residency is the default, and the constraint is baked into how instances are provisioned. For a GDPR review, that's the easy box to tick first.

This is the part most teams don't realise is possible. cachly ships a client-side utility called PIIGuard that masks sensitive values locally, before any request is sent. Emails, phone numbers, IBANs, and your own custom patterns (account numbers, PINs, internal IDs) become deterministic [MASKED_xxxxxx] tokens. The original-to-token mapping stays in memory on your machine and is never transmitted.

import { PIIGuard } from '@cachly-dev/sdk/pii'

const guard = new PIIGuard({
  fields: ['email', 'phone', 'iban'],
  customPatterns: [/\bACC-\d{8}\b/g], // your internal account format
})

const record = {
  summary: 'Refund issued to [email protected] for ACC-40028922',
  amount: 240.0,
}

const safe = guard.maskObject(record)
// summary → "Refund issued to [MASKED_…] for [MASKED_…]"
await cache.set('case:8841', safe)   // server only ever sees masked tokens
guard.clearMapping()                  // drop the local mapping when done

We wrote a full deep-dive on PIIGuard if you want the complete API. The key compliance point: with PIIGuard in your pipeline, the raw PII never appears in any network request — only tokens do.

The whole point of AI memory is that it compounds across your team. Team Brain lets every developer's AI assistant draw on the same shared pool of lessons — the bug someone fixed last month, the migration that went wrong, the pattern your staff engineer always reaches for.

Because the lessons worth sharing are engineering knowledge — "httpOnly cookies fixed the Safari refresh bug", "this migration needs a backfill first" — and because PIIGuard already stripped the sensitive values upstream, your team shares the know-how without anyone's customer data riding along. Knowledge is the asset; the PII never was.

The recommended shape for a regulated team:

• Provision an EU instance (Hetzner DE) — data residency handled.
• Wrap any caching/lesson path that could touch user data in a PIIGuard instance, one per request, with your domain-specific customPatterns.
• Enable Team Brain so lessons compound across engineers — while the masked tokens keep sensitive values out.
• Use brain_hygiene periodically to review and prune what the Brain has stored.

The result is the thing teams assumed they couldn't have: an AI that genuinely remembers your codebase and gets sharper every week — without becoming the weakest link in your data-protection story.