The document-privacy engine you drop into your platform.

DataGuard is an API for entity recognition and document transformation. Send a PDF, DOCX, or spreadsheet; get back the sensitive entities inside it, a policy decision, and a redacted or aliased file, with an audit trail for every step. Enterprise and government grade, running on Sidian's own models, deployable in your cloud or fully air-gapped.

  • API-first
  • White-label
  • On-prem / air-gapped
  • Own models, no third-party AI
  • PII · PHI · Financial · CUI
  • Aligns with NIST 800-171
  • SOC 2 Type II (in progress)

02 / Why an engine, not a regex

Most tools match patterns. DataGuard reads context.

Regex and keyword lists miss what they were never told to look for, and flag what only looks sensitive. DataGuard runs named entity recognition on the actual content, so it catches the SIN in a scanned contract, the diagnosis in a clinical note, and the account number a template never anticipated, across structured and unstructured documents.

Pattern matching

  • Brittle keyword and regex lists you maintain forever
  • Misses novel formats, OCR noise, and context
  • Blocks whole documents, so AI features get switched off

The DataGuard engine

  • NER reads content in context: PII, PHI, financial, CUI
  • Handles messy, scanned, and mixed structured / unstructured data
  • Aliases instead of blocks, so your AI keeps working on safe data

03 / The API

Four calls, one boundary.

Point DataGuard at a document by text, URL, base64 content, or multipart upload. Get entities, a policy decision, or a finished file back. Authenticate with a bearer key; deploy the same API in your own cloud or on-prem.

Base URL

https://api.sidian.io/org/{your-org-slug}

Auth

Authorization: Bearer dg_live_...
POST/scan

Raw entity detection. Returns every entity found, grouped by category. No policy, no mutation.

POST/analyze/review

Full analysis with redaction recommendations. Nothing is changed, so a human can approve first.

POST/analyze/auto

Analyze and redact in one call. Drop a file in, get a redacted file out.

POST/redact

Apply your own rules directly (REDACT or EXCLUDE), no NER step required.

Requestbash
curl https://api.sidian.io/org/acme/scan \
  -H "Authorization: Bearer dg_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "source": {
      "type": "text",
      "text": "Patient Jane Doe, SIN 123-456-789, seen 2026-03-14."
    }
  }'
Responsejson
{
  "entities": [
    { "category": "PII", "type": "PERSON", "text": "Jane Doe",     "confidence": 0.98 },
    { "category": "PII", "type": "SIN",    "text": "123-456-789", "confidence": 0.99 },
    { "category": "PHI", "type": "DATE",   "text": "2026-03-14",  "confidence": 0.91 }
  ]
}

Representative shape. Run it live in the cookbook for the exact payload.

Source types: text, url, content (base64), multipart. File types: PDF, DOCX, PPTX, XLSX, JSON, CSV. Up to 5 GiB per file.

04 / See it work

Identify, transform, comply.

The same three stages every call runs, on one example clinical note. This is a static walkthrough; the live cookbook runs it on your own file.

Source document

Patient Jane Doe (SIN 123-456-789), DOB 1984-02-11, seen at Toronto General on 2026-03-14 for hypertension. Card on file 4111 1111 1111 1111.

EntityCategoryTypeConfidence
Jane DoePIIPERSON0.98
123-456-789PIISIN0.99
1984-02-11PIIDOB0.97
Toronto GeneralPHIFACILITY0.93
hypertensionPHICONDITION0.90
4111 1111 1111 1111FinancialCARD0.99

NER returns every entity with its category and a confidence score.

PARTNER BRIEF·ENTERPRISE-GRADE·ZERO TRUST·v2025.6

05 / What you're integrating

Five controls, one embedded layer.

Drop DataGuard into your product and inherit a complete data-leak-prevention stack: detection, policy, redaction, and a zero trust foundation, held to enterprise and government standards.

6
File formats
4
Entity categories
None
Data sent to outside AI
PII · PHI · FIN · CUI

Entity recognition

NER identifies sensitive data the moment it moves: PII, PHI, financial records, privileged material, and Controlled Unclassified Information, across structured and unstructured content. No manual tagging, no brittle keyword lists.

REDACT · EXCLUDE

Policy engine

A configurable rules engine decides what counts as sensitive and what happens next (redact, alias, flag, or exclude) based on your customer's regulatory regime and your own policies. One engine, many policies, applied consistently everywhere your product sends data.

800-171 / 3.13

Redaction and aliasing

Sensitive content is removed or swapped for safe, consistent stand-ins before data is sent, shared, exported, or passed to an AI model. And because we alias rather than simply block, your AI features keep working on real-shaped data instead of being switched off.

NIST 800-207

Zero trust by design

DataGuard runs on a zero trust model: no implicit trust, least-privilege access, and protection applied at the point of data movement rather than at a perimeter that has already been crossed. Security travels with the data, not just around it.

EVIDENCE

Audit trail

Every detection, rule cited, and action taken is written to an immutable log and shared back to your platform. A defensible record for your customers' regulators and auditors, produced automatically on every call.

In progressAICPA TSC

SOC 2 Type II

We are actively pursuing SOC 2 Type II attestation, holding our own operations to the same standard your enterprise and government customers expect from every vendor in your stack.

DatainEntityrecognitionPolicyengineHumanreviewRedactor aliasSafedata outAudit logcaptures every stage: detection, rule cited, approval, redaction or alias, recipient
Detect, decide, then redact or alias, with the audit log recording every step underneath.

06 / What you can build

One engine, many integration shapes.

Each of these is a real capability mapped to a real endpoint, not a roadmap. Pick the shape that fits your product.

POST /scan

Pre-flight inspection

Before a user shares or uploads a document, show them exactly what is sensitive inside it. Pure detection, nothing is mutated, so you can warn, gate, or route without touching the file.

POST /analyze/auto

Automated redaction pipeline

For trusted, high-volume flows: drop a PDF in, get a redacted PDF out in one call. No human in the loop. Wire it into an export job, an ingestion queue, or an outbound webhook.

/analyze/review + /redact

Human-in-the-loop review queue

Analyze returns recommendations without changing anything, so your reviewers approve or edit first, then commit the redaction. The right shape for legal, medical, and other high-stakes documents.

ALIAS

Keep your AI features on

Alias sensitive values to consistent stand-ins instead of blocking. Your assistant, search, and analytics keep running on real-shaped data, and nothing sensitive ever reaches the model, so you ship AI where competitors have to disable it.

CSV · XLSX · JSON

Sanitize exports and warehouse loads

Clean data on the way out: to a customer export, a BI dashboard, or a data warehouse. Structured formats are first class, so a spreadsheet of records is handled as cleanly as a scanned PDF.

ON-PREM

In-region and air-gapped processing

Run the whole engine inside your own infrastructure, including disconnected networks. Data never leaves your tenancy or crosses a border, and there is no outbound dependency on Sidian or any third-party AI.

07 / How it integrates

Embed it. Don't build it.

DataGuard is API-first and white-label, so it runs inside your product under your brand and your UX. You ship a compliance-grade capability in weeks instead of quarters, and your engineering roadmap stays on your core product. Deploy in your cloud or fully on-premises.

YOUR PLATFORM & DATAYour app and APIsConnected data sourcesYour own data storeDATAGUARDone API boundary1 · Entity recognition2 · Policy engine3 · Human review4 · Redact or aliasRECIPIENTSPartners · RecipientsRegulators · AuditorsAI models · Exportsthey receive only the safe versionAudit log: a defensible record of every decision, shared back to your platform
DataGuard sits between your data and the outside world, returning safe data and a complete audit trail.

MODEL 01

API

Call DataGuard from your product and keep your own UI. The fastest path to a working integration.

MODEL 02

Embedded / white-label

DataGuard runs inside your platform under your brand. Your customers never leave your product.

MODEL 03

On-prem / air-gapped

Deploy entirely inside your own infrastructure, including disconnected and air-gapped networks. No outbound dependency required.

08 / Why teams build on DataGuard

Embed a capability, not a cost center.

You bring the platform and the reach. DataGuard brings the engine and the public-sector credibility, already built and already vetted.

Ship in weeks, not quarters

API-first and white-label, so DataGuard runs inside your product under your brand and your UX. You get a compliance-grade capability without pulling your engineers off your core roadmap.

Clear security reviews faster

Our controls map to the frameworks your customers operate under, so integrating DataGuard helps both of you get through security questionnaires and reviews with evidence instead of promises.

Differentiate with safe AI

Offer the AI features your competitors have to block, because aliasing keeps sensitive data out of the model while keeping the output useful.

Lower risk, raise retention

A defensible audit trail reduces breach and data-processing liability for you and your customers, and an embedded compliance layer is deep, sticky, and grows account value over time.

09 / Security & data sovereignty

Built for the most sensitive environments.

For government, defense, and regulated work, where the data sits and what touches it matters as much as the result. DataGuard is designed so neither ever leaves your control.

Data residency

In plain terms

Your data stays in your environment and your jurisdiction. Nothing is copied out to be processed.

Under the hood

Deployed in-region; data never leaves your tenancy or crosses borders during processing.

Our own models do the work

In plain terms

Protection is run by Sidian's own purpose-built models. No third-party AI ever sees your data.

Under the hood

No calls to OpenAI, Anthropic, or any external API. Models are trained and hosted by Sidian on real regulated documents.

On-prem deployable

In plain terms

Run DataGuard entirely inside your own infrastructure, including air-gapped and disconnected environments.

Under the hood

Containerized into your cloud, private cloud, or on-prem / air-gapped network. No outbound dependency required.

10 / Trust & assurance

Held to the standard your customers audit against.

The engine is government-proven, and Sidian carries the attestations and coverage your enterprise buyers expect from a vendor in their stack.

In progress

SOC 2 Type II

Actively pursuing SOC 2 Type II attestation, so your security team can point to an independent report rather than take our word for it.

Cyber liability insurance

Sidian carries cyber liability insurance, so the risk you take on by integrating us is backed, not just asserted.

Government-proven engine

Our core IP was procured and proven through a Canadian public-safety procurement and aligns with NIST SP 800-171, CMMC, and FedRAMP control families for handling Controlled Unclassified Information.

  • SOC 2 Type II (in progress)
  • Cyber liability insured
  • Government-proven
  • NIST SP 800-171
  • Zero trust (NIST 800-207)
  • CUI ready

11 / Let's build together

Give your product a government-ready data layer.

Without building it yourself. We bring the engine and the public-sector credibility; you bring the platform and the reach; your customers get to move fast without tripping a compliance wire.

BBen ReichweinFounder & CEO, Sidian

Talk to our team

Bring your platform and a real use case. In one short call we'll map exactly where DataGuard fits and what an integration looks like.