Redact

Strip PII before it reaches your LLM.

Redact sits between your app and any LLM as a drop-in proxy. It removes names, emails, phone numbers, addresses and custom patterns before requests leave your infrastructure, then restores them in the response.

Join the waitlist

Your app sends

Patient Jane Doe, jane.doe@example.com, 555-0100, reports chest pain since Monday.

Your LLM receives

Patient [NAME_1], [EMAIL_1], [PHONE_1], reports chest pain since Monday.

Fictional sample data.

What you get

Drop-in proxy

Point your LLM client at Redact instead of the provider. No SDK to adopt, no prompt code to rewrite.

Round-trip restore

Placeholders go out, real values come back. Your users read a complete answer; the model never saw the data.

Zero PII storage

Values are held only for the life of the request, then dropped. Nothing to breach, nothing to purge.

Per-request reports

Every call returns what was detected and replaced: evidence you can hand to a customer’s security review.

How it works

  1. 01

    Send the request to Redact

    Your app calls Redact with the same payload it sends your LLM today. One base URL changes.

  2. 02

    Detect and redact

    Names, emails, phone numbers, addresses and your custom patterns become typed placeholders inside your infrastructure.

  3. 03

    Forward, restore, report

    The clean prompt goes to any LLM. Values are restored in the response, and a redaction report comes with it.

Vertical compliance packs

Pre-built detectors tuned for the data each regulated vertical actually handles. Turn one on, add your own patterns on top.

Healthcare

Patient triage

Tuned for patient names, dates of birth, contact details and record identifiers in symptom intake and triage flows.

Legal

Contract review

Tuned for party names, signatories, addresses and account details buried in clauses and schedules.

HR

HR chatbots

Tuned for employee names, personal contacts and identifiers that surface in policy and leave questions.

Why Redact

Built for teams who have to show, not just say, that patient, client and employee data stays out of model prompts.

No SDK rewrite
Redact is a proxy, not a library you thread through every call site.
Restores, not just strips
Redaction that breaks the answer is not usable. Redact puts the values back.
Audit evidence per request
A report on every call, not a dashboard you screenshot once a quarter.
Nothing stored
Zero PII storage by design, so Redact never becomes the new data store to protect.
Your patterns too
Add custom patterns for internal IDs, case numbers or anything your domain uses.
Vertical packs
Start with detectors pre-tuned for healthcare, legal and HR instead of from zero.

Questions

Does Redact store PII?

No. Detected values are kept in memory only long enough to restore them in the response, then discarded. Reports describe what was redacted without keeping the values.

Which LLMs does it work with?

Redact is designed to sit in front of any LLM provider your app calls over an API. Tell us which ones you use when you join the waitlist.

How do I integrate it?

Change your LLM client’s base URL to Redact. Your prompts, models and response handling stay the same.

Can I add custom patterns?

Yes. Define patterns for identifiers specific to your product, such as internal patient or case numbers, alongside the built-in detectors.

What is in a redaction report?

For each request: which entity types were found, how many, and which placeholders replaced them. Useful as evidence in security reviews.

Can it run in my own infrastructure?

Redact is built so that redaction happens before requests leave your infrastructure. Deployment options for early access are shared with the waitlist.

Ship the LLM feature. Keep the PII.

Usage-based plans. Join the waitlist for early access and early-access pricing.