Anonymizer

What it is

Anonymizer is a pre-processing layer for prompts. Before a request is sent to a frontier model, it detects and replaces sensitive entities — names, emails, phone numbers, national IDs and bank details — with neutral placeholders.

The model still understands the structure and intent of the request, but never sees the real data.

How it works

• Detection — a lightweight model (or a rule engine, depending on the case) locates sensitive spans in the text.
• Redaction — each entity is replaced with a typed placeholder such as [NAME_1] or [EMAIL_2].
• Re-identification — optionally, placeholders in the model’s response are mapped back to the original values on the client side.

Why it matters

Many teams want the capability of frontier models without the risk of leaking customer data into a third-party service. Anonymizer lets you keep the model on the outside and the data on the inside.

Try the redaction pipeline below — it runs locally against a mock model.