Redwood · PII redaction for public records

Documents go in. A redacted record comes out. Invisible to everyone else.

Our in-house model, WipeWay v3, finds the personal information in case files and records requests. Your reviewer decides every proposal — one at a time or in bulk — the page is blacked out for real, and every decision is logged. The service runs on infrastructure operated by Cloudless Labs; we never access it. Our commitment: no one at Cloudless Labs sees your documents, and nothing you submit trains anything.

90% fewer false detections than the open model it's built on — with nothing missed on the held-out pages.1

Request a pilotSee how it works
Diagram. Your documents enter a black box run by Cloudless Labs. Inside it: detection, review, blackout, and an append-only audit trail. What comes out: redacted documents with a comprehensive audit log. No one at Cloudless Labs sees your documents, nothing is used for training, and the redaction path calls no outside service.
InBulk drop your documents

Inside the black box · run by Cloudless Labs · unseen by us

  1. Detect
  2. Review
  3. Redact
  4. Log
  • Seen by anyone at Cloudless Labsnothing
  • Used for trainingnothing
  • Outside services in the redaction pathnone
OutRedacted documents with a comprehensive audit log

Measured results

What happens inside the box.

Six steps, none of them seen by anyone at Cloudless Labs. Every step leaves a record.

  1. In

    Bulk drop your documents — a whole batch at once, uploaded or dropped in a folder.

  2. Detect

    Ten PII types. Every detection becomes a proposal with its type and the matched text.

  3. Review

    A reviewer accepts, rejects or draws a box — from the keyboard, singly or in bulk. Export waits until every page is marked done.

  4. Redact

    Pixels inside every approved box are blacked. The page is rebuilt from the images, and the text is re-read afterwards.

  5. Log

    An append-only audit trail, enforced by the database: updates and deletes are refused. A SHA-256 of the output is recorded at export.

  6. Out

    Redacted documents with a comprehensive audit log. The log records positions and decisions, never the redacted text.

Skip the demonstration

PII identified in seconds, so human review is fast.

  1. In

    Bulk drop your documents — a whole batch at once, uploaded or dropped in a folder.

  2. Detect

    Ten PII types. Every detection becomes a proposal with its type and the matched text.

  3. Review

    A reviewer accepts, rejects or draws a box — from the keyboard, singly or in bulk. Export waits until every page is marked done.

  4. Redact

    Pixels inside every approved box are blacked. The page is rebuilt from the images, and the text is re-read afterwards.

  5. Log

    An append-only audit trail, enforced by the database: updates and deletes are refused. A SHA-256 of the output is recorded at export.

  6. Out

    Redacted documents with a comprehensive audit log. The log records positions and decisions, never the redacted text.

Request a pilot

INCIDENT / OFFENSE REPORT

Case No. [redacted: case number] CH-2026-004417Report copyRedacted copy

Incident type
Fraud — unauthorized use of a financial card
Date reported
[detected, released by policy: date] 03/02/2026detected · released by policy
Date of incident
[detected, released by policy: date] 02/27/2026
Location of incident
[redacted: full address] 2210 Millrace Rd, Unit 4, Cedar Hollow, OR 97031
Reporting officer
[redacted: person name] Jared Okafor, Badge 2231

COMPLAINANT / VICTIM

Name
[redacted: person name] Priya Raman
Date of birth
[redacted: date of birth] 04/19/1988
SSN
[redacted: social security number] 000-61-2270
Home address
[redacted: full address] 77 Quarry Hill Rd, Apt 3B, Cedar Hollow, OR 97031
Phone
[redacted: phone number] (503) 555-0177
Email
Employer
[redacted: organization] Harbor Point Property Management

WITNESS

Name
[redacted: person name] Tomas Ferreira
Phone
[redacted: phone number] (503) 555-0163
Relationship
Coworker of the complainant

PROPERTY / FINANCIAL

Instrument
Debit card, number [redacted: credit card number] 4111 1111 1111 1111
Issuer
[redacted: organization] Larkspur Community Credit Union
Reported loss
$1,284.50

NARRATIVE

On [detected, released by policy: date] 02/27/2026 at approximately 14:20, [redacted: person name] Priya Raman received a text alert from [redacted: organization] Larkspur Community Credit Union for two purchases she stated she did not make. She reported that the card had not left her possession and that she had last used it on [detected, released by policy: date] 02/25/2026 at a fuel station near her residence. Witness [redacted: person name] Tomas Ferreira, a coworker at [redacted: organization] Harbor Point Property Management, stated that he saw an unknown male photograph the complainant's card while it was on her desk on [detected, released by policy: date] 02/26/2026. The complainant was advised to contact the issuer to dispute the charges and was provided case number [redacted: case number] CH-2026-004417 for reference. Follow-up assigned to Detective [redacted: person name] Rosa Castellano.

Reporting officer
[redacted: person name] Jared Okafor, Badge 2231
Approved by
Sgt. [redacted: person name] Lena Marsh, Badge 1108
Date approved
[detected, released by policy: date] 03/03/2026

Page 1 of 3Sample · all data fictitious

A black box, by design.

Redwood runs as a service on systems Cloudless Labs operates. Our commitment to you: no one at Cloudless Labs sees your documents — not during detection, not during review, not after export — nothing you submit trains anything, and everything is computed in house, never handed to a third-party service. Its redaction logs and audit exports name what was found, so keep them under the same controls as the records themselves.

  • Nothing to install, nothing to host

    No software to run and no hardware to size. Your reviewers work in a browser; everything else is computed in house, on systems Cloudless Labs operates.

  • No outside services — enforced, then tested

    The redaction path calls out to nothing: offline mode is forced when the software loads, a regression test asserts that nothing can re-enable it, and the built review screen is checked for outbound references. Everything is computed in house: your documents are handled by Redwood alone, never handed to a third-party service.

  • A better model. Nothing learned from you.

    WipeWay is an improved version of an open model (Apache-2.0 lineage). There is no training code in the product and no feedback loop from reviewer decisions — nothing you submit is used to train anything.

Prove it on your own records.

A pilot runs Redwood — the whole workflow — on a sample of your records, through the same black box, and gives you the same scorecard we publish: recall and false detections, per PII type, on your documents.