The Privacy
Gateway

Prompts leave clean. Responses arrive real.

1.1 PII Vault Live

Every prompt is intercepted. Personal data is replaced with reversible tokens before it leaves your network. The AI sees placeholders. After it replies, Skilmo maps the originals back in.

Write your text or use this sample:

1.2 Skill Vault Q3 2026

A second vault, this time of instructions. Any skill or tool a model loads — a calendar export, a court-precedents lookup, a tax-rate table — must be reviewed and Ed25519-signed by Skilmo first. Unsigned bundles never reach the model.

IT sets this up once. End users never know it's there.

Set up on your laptop or local server in minutes — no cloud configuration, no data leaving your network.

Skilmo ships as a single signed container. It bundles multiple PII classifiers, a local NER model, and a small on-device LLM. The initial download is 4–7 GB and runs entirely offline from that point forward.

1. Pull the container: docker pull skilmo/pii-vault:latest
2. Run it once — a guided setup wizard launches automatically
3. When prompted, enter your Skills Registry account ID to activate

Minimal configuration, data is safe in your infra:

{
  "mcpServers": {
    "skilmo-pii": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-v", "skilmo-audit:/var/log/pii-vault",
        "skilmo/pii-vault:latest"
      ]
    }
  }
}

Standard PII

Email, phone, credit card, IBAN, IP address, national ID, date of birth, passport, URL, credentials.

Baltic & Nordic

Estonian isikukood (mod-11 validated), Latvian personas kods (2021 format), Lithuanian asmens kodas, Finnish henkilötunnus, Estonian company registration number, Estonian IBAN (EE prefix).

Organisation-specific

Employee IDs, project codenames, client references — defined once via company.yaml, detected automatically on every request.

Two modes: fast (regex pipeline, <5ms) and accurate(NER + ML, <300ms).

Every sanitise operation produces a structured log entry:

2026-05-18T09:14:02Z | session:s_a3f9 | op:sanitize | entities:3 | PERSON_1=Marta Tamm[0.97,EstBERT_NER] | EMAIL_ADDRESS_1=marta.tamm@example.ee[1.0,EmailRecognizer] | EE_PERSONAL_CODE_1=49002128026[1.0,EEPersonalCodeRecognizer]

Exportable as JSON or CSV. Covers GDPR Article 30 records of processing. Ready for DPO review and regulatory audit.

Skilmo is incorporated in Estonia (EU). A standard Data Processing Agreement is included with every paid plan. ISO 27001 deployment review available on request.

Microsoft Presidio is what we build on. It is excellent. It does not do these things:

1. Detect Estonian, Latvian, or Lithuanian personal identifiers.
2. Provide a caller-carries-key API for reversible anonymisation.
3. Expose MCP tools for Claude Code, Cursor, or Copilot.
4. Write a GDPR-aligned audit log with recognizer metadata.
5. Ship as a single zero-outbound Docker container.
6. Come with a DPA and a human auditor who signs off.

We are the operational layer Presidio needs to work in a regulated European company.

What is PII (personally identifiable information)?

PII is any data that can identify a specific individual — name, email address, phone number, national ID, date of birth, passport number, financial account numbers, IP address, or any combination that makes a person identifiable. GDPR Recital 4 defines it as any information that directly or indirectly identifies a natural person.

What is reversible anonymisation?

Skilmo replaces each PII span in your text with a placeholder token such as [PERSON_1] or [EMAIL_ADDRESS_1]. The original values are stored locally under a one-time key. After your AI model responds, Skilmo substitutes the originals back into the response. The net effect: the AI never sees the real values, but your application receives a response with the originals restored.

What is MCP (Model Context Protocol)?

MCP is an open standard published by Anthropic for connecting AI assistants to external tools and data sources. Claude Code, Cursor, and Copilot can call MCP-compatible tools directly during a conversation. Skilmo exposes its PII sanitiser as two MCP tools — sanitize_pii and restore_pii — so AI coding assistants can redact and restore personal data without leaving the IDE.

What is Skilmo?

Skilmo is a local-first PII sanitiser for companies using Claude, GPT, or Gemini. It runs in Docker on a laptop or a server. Nothing leaves the building.

What PII does Skilmo detect?

Standard PII (email, phone, credit card, IBAN, IP address, national ID, date of birth, passport, URL, credentials), Baltic & Nordic identifiers (Estonian isikukood with mod-11 validation, Latvian personas kods, Lithuanian asmens kodas, Finnish henkilötunnus, Estonian company registration, Estonian IBAN), and organisation-specific identifiers (employee IDs, project codenames, client references) defined via company.yaml.

How is Skilmo different from Microsoft Presidio?

Skilmo builds on Presidio and adds: Baltic & Nordic personal identifier detection, a caller-carries-key API for reversible anonymisation, MCP tools for Claude Code, a GDPR-aligned audit log with recognizer metadata, a single zero-outbound Docker container, and a DPA backed by a human auditor.

Where is data sent?

Nowhere. Skilmo runs entirely inside your Docker container. Sanitised text flows out to your chosen AI provider; original PII is restored locally in the response.

Is Skilmo GDPR-compliant?

Skilmo helps you meet GDPR Articles 25, 30, and 32. Every sanitise operation produces a structured audit log entry; logs are exportable as JSON or CSV. Infopol OÜ is incorporated in Estonia (EU) and provides a standard Data Processing Agreement with every paid plan.

Klaus Jõgi — ISO 27001 Lead Auditor and Consultant

“Meaningful compliance requires understanding business, not just filling tables. We build security programs that work in the real world.”

20 years in IT and information security. ISO 27001:2022 Lead Auditor. Building something similar? Need Estonian or Baltic language coverage for a production system? Testing compliance for a specific EU regulation?

klaus.jogi@skilmo.com  ·  +372 5886 6888  ·  linkedin.com/in/klausjogi

Infopol OÜ — Tallinn, Estonia  ·  Registrikood 16494599  ·  KMKR EE102969308