Every company that starts using AI in earnest runs into the same dilemma: the most powerful models live outside — in the clouds of Anthropic, OpenAI, Google and other providers — and your data includes names, national ID numbers, payroll records, customer histories. Do you give up the best models, or cross your fingers on compliance? We believe the dilemma is a false one, and we solve it with two pieces that we use in everything we build.
Piece 1: anonymise before sending, restore on the way back
Our anonymiser agent sits between the applications and the AI models. Before a single piece of data leaves for an external provider:
- It detects personal information in text — and also in scanned PDFs, via OCR — including Spanish identifiers: DNI, NIE, IBAN, social security numbers.
- It replaces each piece of data with a neutral marker, so the model works with the full document but never knows who it's talking about.
- When the response comes back, it restores the original data in place. For the user, the process is invisible.
The practical result: you can use any model on the market — including the most powerful from each provider — because personal information never leaves your perimeter. GDPR stops being a constraint and becomes a property of the system.
It's not about trusting the provider to handle your data well: it's about them never having it.
Piece 2: every query, to the model it deserves
The second piece answers another question we're always asked: which model should I use? The honest answer is that there isn't a single best one — there's a best one for each type of task, and it changes every few months.
NubemRouter, our multi-provider panel, solves this at the root: it knows every provider on the market and routes each request to the model with the best quality-to-price ratio for that type of query. A simple classification doesn't need the most expensive model; a complex legal analysis does. The routing is decided on every call, with fallback mechanisms if a provider fails, and with spend visible on a dashboard — no surprises at the end of the month.
Together, they change the conversation
With these two pieces, the question "can we use AI with our data?" becomes "which process do we want to improve first?". Security and cost stop being objections and become configuration: data travels anonymised, every task uses the optimal model, and everything is logged.
This is how we work under the hood across all our agents — from due diligence to document OCR. If your company has held back on AI because of GDPR or runaway costs, this is exactly the problem these two pieces solve.