Redact
Sensitive data, masked at the source.
Find and mask names, account numbers, addresses and anything else your policy calls sensitive, before a document is stored, shared or sent to another system. The model learns your definition of sensitive, not a generic list of patterns.
note_2291.txt · before it's stored
Spoke with NAME about the card ending ACCOUNT. She moved to ADDRESS and her new number is PHONE. Dispute DSP-30918 stays open.
What it does
Built for tasks like these.
- De-identify records before analytics
- Scrub transcripts before they are stored
- Mask PII before text reaches an outside API
- Prepare documents for outside review
How it works
From your examples to a model you own.
- 01
Mark what counts as sensitive
Highlight it in a few hundred real documents: the obvious fields and the ones only your team would know.
- 02
Train the redactor
The model learns your categories and formats, on a machine next to the data.
- 03
Redact at the source
Run it before storage, sharing or any outside call. Only masked text moves on.
Private by default
Your records never leave the building.
A redaction service in someone else's cloud has to see the raw data to hide it. Running the redactor on your own hardware is the only way the unredacted text never leaves.
- Your customers' data leaves the building on every call
- The model changes whenever they ship
- You pay for a huge general model to do one narrow task
- The request never leaves the device
- The model is a file you own; it changes when you retrain it
- Small enough to run on hardware you already have
Trained, not prompted
Accurate on your task.
Patterns catch the obvious formats and miss the rest: a name typed in lowercase, an account number split across lines, an address written out in words. A trained model learns what sensitive looks like in your documents, and keeps the text around it exactly as written.
InputSpoke w/ maria gonzalez re: card ending 4471, she's at 18 Elm St apt 2
General model, prompted
Spoke with [REDACTED] regarding a card account. She lives at [REDACTED].
Micromodel, trained on your examples
Spoke w/ [NAME] re: card ending [ACCOUNT], she's at [ADDRESS]
Illustrative example.
And it costs less
No per-call bill.
Redaction runs on every document before anything else can. Per-call pricing on that step taxes your whole pipeline; a local model doesn't.
$120,000 annual savings
See it for yourself
Run this model on your own CPU.
A guided demo, about a minute. It downloads once and runs entirely in your browser; nothing you type is sent anywhere.Open the demoRead and write
Related solutions.
Have a task like this?
We'll tell you if a micromodel can do it. Or book a demo, or email hello@micromodel.company.