Redact sensitive data. Keep control.

Semred combines context-aware redaction with an inspectable safety protocol—from the first rule pass to the final export. Remove PHI, PII, and confidential detail while preserving meaning.

Explore the safety protocol

Rules run on your host · External AI off by default · Leak-checked exports

Safety protocol

Your document has one way out, and you hold it.

The reason teams cannot use a general‑purpose model is the reason Semred exists. Four guarantees hold on every run. See the full protocol

  • Rules run on your host, first Deterministic detection with no network call.
  • External AI is off by default Only an administrator can turn it on, and the run records that it ran.
  • A model never sees the original If enabled, it receives text the rules have already redacted.
  • Every export is leak‑checked Read back and compared with the records before it downloads.
Safety protocol

Where your document goes. And where it never does.

Redaction exists because the document cannot be sent anywhere. So Semred draws the boundary first and works inside it. Watch one document make the trip.

  1. Your documentImported into a private workspace on your host.
  2. Rule passDeterministic detection. Runs here, no network call.
  3. Your reviewEvery replacement listed and inspected before export.
  4. Off by default External modelOpens only when an administrator sets the flag. Receives rule‑redacted text, never the original.
  5. Leak‑checked exportRead back and compared with the records before it downloads.

Your host · nothing leaves without the flag

  1. 01
    The rule pass runs on your host, first.

    Deterministic detection runs before anything else and never makes a network call.

    Enforced in code
  2. 02
    External AI is off until an administrator turns it on.

    The semantic pass only runs when the deployment sets SEMRED_ALLOW_EXTERNAL_LLM=true. Each run records whether it did.

    Admin setting
  3. 03
    A model, if enabled, sees rule‑redacted text. Never your original.

    Stage two receives the stage‑one output, with names, identifiers and dates already replaced.

    Enforced in code
  4. 04
    Logs hold no content.

    No source text, prompts, model output or replacement mappings are written to service logs.

    Enforced in code
  5. 05
    Every export is read back and checked for leaks.

    TXT, DOCX and PDF exports are re‑extracted and compared against the records before they download. Residual data blocks the export.

    Enforced in code
  6. 06
    Every run carries provenance, without content.

    Policy fingerprint, detector and prompt versions, model name and validation state ship with each result.

    Enforced in code
  7. 07
    What we do not claim.

    PDF export writes a new text PDF. It does not scrub hidden layers, metadata or embedded images of an original PDF.

    Stated limit

Deployment decisions that stay with you, such as identity provider, encrypted storage and retention, are listed in the security runbook we share with design partners. Discuss a deployment

Redaction lifecycle

A staged, inspectable redaction lifecycle.

Semred separates deterministic coverage from contextual review. Each applied change is inspectable, and the run records which stages actually completed.

Detect

Deterministic rules run first. When enabled, an LLM reviews the partially redacted text for contextual or indirect identifiers, and every proposed action is validated before application.

Canonicalize

Overlapping and duplicate spans are normalized within each stage, with stable labels and conservative conflict handling before that stage rewrites text.

Rewrite

Accepted spans are replaced locally with stable placeholders. Unchanged source text is preserved rather than silently regenerated.

Score

A similarity score compares the source and result. Embeddings are used when enabled and available; otherwise a deterministic lexical fallback is used. It is not a safety or coverage score.

Audit

Every applied replacement records its type, source value, replacement, stage, and review metadata. Audit trails can be exported as JSON or CSV.

Context, not regex

It knows what the word is doing.

A pattern matcher sees nine digits. Semred sees whether those digits are an SSN, a case number, or a quantity — and whether the sentence still works once they’re gone.

Talk to us about your corpus
  • Person Bob Smith → [Patient A]
  • Org Johns Hopkins → [Hospital]
  • Date March 12, 2024 → [Date]
  • Condition cancer → [Condition]

Replacement records · inspector view

Similarity scoring

Measure what survived.

Redaction that destroys meaning deserves review. Semred compares the source and result so you can spot substantial change; the score is a similarity signal, not proof of safe coverage.

What the score does not mean
Discharge summary0.94
Referral letter0.91
Vendor contract0.88
Support transcript0.72

Illustrative sample. Similarity describes meaning, not coverage.

Audit trail

Every change, on the record.

Each replacement carries its type, original value, substitute, and reason. Export the whole trail as JSON or CSV when compliance asks how a document got that way.

Where the data goes, and where it never does
14:02:11 PERSON replaced · deterministic stage
14:02:11 ORG proposed · semantic review accepted
14:02:12 DATE replaced · rule matched
14:02:12 Export blocked · source edited after last run

Audit records · JSON and CSV export

Solutions

For teams that can’t leak.

Semred is designed for environments where a single exposed record is unacceptable.

Healthcare

De‑identify clinical notes, referrals, and messages while preserving the clinical detail that makes them useful downstream.

Legal

Sanitize contracts, discovery documents, and briefs before they reach vendors, co‑counsel, or a model.

Enterprise

Strip internal docs, tickets, and logs of identifying detail before they leave your environment.

Stated plainly

What Semred does not claim.

Redaction tools that promise completeness are the ones that get people hurt. Here is where our guarantees stop.

A high score is not safety

Semantic similarity describes how much meaning survived. It says nothing about whether every sensitive span was found.

Zero records is not a clean document

An empty replacement list means nothing matched — not that there was nothing to match. Review still applies.

Review stays in the loop

Semred is built to make human review fast and evidence‑backed, not to remove it. Spans the model alone proposed are flagged as such.

Have documents you can’t send anywhere?

That’s the case we built this for. Tell us what you’re sitting on and we’ll show you what comes out the other side.

Request access
Pricing

Start with a focused pilot.

We work directly with teams that have real redaction problems — not toy demos. No self‑serve tier while we’re shaping the product around design partners.

Design partner

Work directly with us to shape Semred around your workflows.

  • Pipeline tuned to your document types and entity taxonomy
  • Private deployment in your environment
  • Direct line to the people building it
  • Pricing agreed after the pilot, not before
Request access
Contact

Talk to Semred.

Share a bit about your use case and we’ll get in touch.