IN-HOUSE SYSTEM

Private data stays private: local AI first, the cloud only when it's safe.

Context

We use AI across our own operations every day, and some of what we handle should never leave our own hardware. So we built the routing we'd want a client to have.

What we built

  • A redaction step that strips names, contact details and identifiers before any text leaves our systems.
  • Requests routed by sensitivity: private work runs on local models, and only cleaned text ever reaches a cloud model.
  • A spending cap on every cloud key, so a runaway loop can't run up a bill.

Why it matters

It's the pattern we bring to clients who handle health, legal or financial records.

Stack

n8n · local models · Python

Start with an audit.