RFC v1.0 SPECIFICATION · OPEN CONSORTIUM · ZERO TRUST DATA SANITIZATION (ZTDS™)

ZTDS^™ (Zero-Trust Data Sanitization) The Open Standard & Enterprise AI Trust Seal

Q: How does ZTDS protect sensitive data in Model Context Protocol (MCP) for Cursor and Claude Desktop?

The open reference MCP server ( ztds-mcp ) executes locally on your workstation as a child process over stdio JSON-RPC pipes. It intercepts prompts, code diffs, and tool arguments in volatile RAM before network transmission. Cleartext API keys, internal IPs, and customer PII are replaced with contextual surrogate tokens ( [KEY_1] , [EMAIL_1] ) so AI reasoning is unaffected, while zero bytes of sensitive data leave the machine. The mapping table is preserved strictly in client volatile memory and re-applied locally to responses before display.

Q: How does ZTDS solve GDPR Article 17 Right to be Forgotten in RAG vector databases?

In standard RAG architectures, raw customer PII is embedded into dense vectors stored in Pinecone, Qdrant, or Milvus. When a customer exercises their GDPR Article 17 "Right to be Forgotten", dense embeddings cannot be deleted individually without re-indexing the entire multi-million vector index. ZTDS sanitizes raw entities into deterministic surrogate tokens prior to embedding generation . The vector database stores mathematical embeddings of de-identified tokens, ensuring zero direct linkability to natural persons. Deleting the customer's reversible token map from local keystores renders the vectors permanently de-identified without needing a costly corpus re-index.

Q: How does ZTDS eliminate data processor liability under GDPR Article 28?

Under GDPR Article 28, a Data Processing Agreement (DPA) is required when external third parties process personal data on your behalf. Because the ZTDS protocol executes 100% inside client volatile host RAM with 0.00 bytes of raw personal data transmitted over the network, ZTDS software never receives personal data and is physically eliminated from the data processor supply chain. By neutralizing direct PII in local RAM and reducing entropy across vertical profiles, ZTDS establishes a proactive technical safe harbor under GDPR Article 32 (Technical & Organizational Measures) and Israel Protection of Privacy Law Amendment 13 Section 17, drastically mitigating statutory liability and eliminating subprocessor risk assessments.

Q: What is the difference between open reference ztds-mcp and @privacyscrubber/mcp-server?

ztds-mcp is the open-source reference implementation published under Apache 2.0 by the ZTDS Consortium, providing baseline universal entity sanitization (emails, keys, credit cards, SSNs, IPs) for individual developers. @privacyscrubber/mcp-server is the certified enterprise production engine engineered by BrandMeWeb, including 30 specialized industry profiles (HIPAA clinical PHI, banking routing/IBANs, legal work product, cloud IAM credentials) and offline Ed25519 cryptographic token licensing for corporate teams.

Q: How can I independently verify zero external network egress in my own environment?

ZTDS conforms to the Airplane Mode Verification Protocol : load the web playground or run the CLI auditor npx ztds-audit --strict , physically disconnect your network interface (or enable Airplane Mode), and observe that all sanitization, detokenization, and verification execute with 100% functionality without error. You can also run Wireshark or inspect the Chrome DevTools Network panel to confirm that exactly 0 network requests are dispatched during processing.

Airplane-Mode Privacy for AI. Verifiable proof that enterprise prompts, customer records, and trade secrets never leak to external clouds. The open RFC v1.0 standard and cryptographic accreditation authority that clears DPO and InfoSec reviews with 0.00 bytes data egress.

Physical In-RAM Execution Topology

Zero Backend Architecture · 0 ZTDS Servers in Loop

1. Host Process

User Application

Sensitive prompt entered in browser, local app, or IDE.

2. In-RAM Enclave < 0.8 ms

Local Volatile RAM

Cleartext swapped for context-preserving tokens. No disk write.

3. Network Socket

Encrypted Transport

Only surrogate tokens cross the NIC. Zero raw PII egress.

4. External Cloud

Public LLM API

Inference executes on tokens. Provider sees 0.00 B private data.

Role-Based Fast Track Navigation

CISO · Infosec →

CISO Procurement & Safe Harbor Pack

GDPR Art. 32 Safe Harbor, Art. 28 processor path elimination, SOC 2 Type II evidence telemetry, and CISO ROI model.

View CISO Dossier →

AI Engineers · MCP →

In-RAM MCP & SDK Architecture

Model Context Protocol guardrails for Claude & Cursor, zero-egress pipes, and LangChain blueprints.

Explore MCP Spec →

Legal · Compliance →

Evidentiary Sovereignty

Forensic chain of custody under FRE 502 & 902, GDPR Recital 26 de-identification, and patent covenants.

Review Legal Treatises →

Quick test:

Apply for Accreditation → Verify Conformance & A4 PDF → CISO Procurement Pack →

✓ 100% Free Open Specification (Apache 2.0 / CC BY 4.0) · ✓ Ed25519 Cryptographic Attestation · ✓ GDPR Art. 32 Safe Harbor & Processor Path Elimination · Enterprise B2B Velocity ↓ · Calculate CISO ROI →

0.00 B

Zero Network Egress

Hardware In-RAM boundary physically prevents sensitive data from crossing network sockets.

Ed25519

Cryptographic Proof

Deterministic RFC 8032 digital signatures and verifiable A4 Conformance Certificates.

RFC v1.0

Open Standard

Vendor-neutral open consortium under Apache 2.0 / CC BY 4.0 on IETF track.

Art. 32

Statutory Safe Harbor

GDPR Art. 32 technical due diligence; eliminates GDPR Art. 28 processor liability.

* Conformance claims apply when implementations satisfy all 4 ZTDS RFC v1.0 invariants as verified by the Perimeter Scanner and validated via Cryptographic Conformance Receipts. This does not constitute legal counsel.

BUSINESS IMPACT · THE TWO SIDES OF ZTDS ACCREDITATION

How ZTDS Certification Accelerates B2B Enterprise AI Sales

Enterprise procurement stalls when AI prompts touch cloud vendors. ZTDS replaces multi-month legal negotiations and invasive InfoSec audits with deterministic, verifiable in-RAM security proofs.

For AI SaaS & Solution Providers Sales Velocity

Unlock Enterprise B2B Sales · Turn InfoSec Objections into Instant Sign-offs

When enterprise prospects refuse to share client data with frontier LLMs, deals stall in InfoSec and Legal review for 3 to 6 months. By embedding a ZTDS-certified sanitization layer, you can prove zero cleartext egress to external model hosts.

✓ Instant Security Questionnaire Sign-Off: Provide an official A4 Certificate of Conformance and Ed25519 cryptographic receipts.

✓ Zero Subprocessor Friction: Eliminate complex multi-party Data Processing Agreements (GDPR Art. 28) for AI model providers.

✓ Live Verification Trust Seal: Showcase the ZTDS Certified Product Trust Badge on your site and customer RFP responses.

Apply for Product Accreditation → Explore Certified Registry →

For CISOs, Legal Counsel & DPOs Statutory Shield

Statutory Risk Minimization & Safe Harbor · Zero Personal Officer Liability

Allow internal teams to safely leverage frontier AI models without risking corporate data leaks, regulatory investigation, or officer liability under strict privacy statutes.

✓ GDPR Art. 32 Technical Safeguards: Provable defense-in-depth proving proactive statutory due diligence against corporate administrative fines.

✓ Israel Protection of Privacy Amendment 13 (Sec. 17): Shields corporate directors and DPOs from personal liability through documented technical controls.

✓ Deterministic Cryptographic Attestation: Generate automated audit receipts proving zero bytes of raw PII left internal memory boundaries.

Download CISO Procurement Pack → Inspect A4 Conformance Certificate →

B2B Sales Velocity Benchmark

The Enterprise AI Adoption Chasm: 6 Months vs. 48 Hours

90% Deal Friction Eliminated

Traditional AI Procurement 3 – 6 Months

× Invasive InfoSec Review: 200+ vendor questionnaire items on cloud data handling, cloud tenancy, and multi-tenant isolation.

× Subprocessor Legal Deadlock: Endless DPA redlines (GDPR Art. 28, SCCs, Schrems II cross-border risk) for frontier model hosts.

× Personal Officer Exposure: CISO and DPO hesitate under statutory liability (GDPR Art. 83 / Israel PPL Amendment 13).

Result: Deal stalls in risk committee Lost ACV

ZTDS-Certified AI Adoption 48 Hours

✓ Turnkey Compliance Evidence: Hand customer InfoSec an official A4 Certificate of Conformance + offline Ed25519 cryptographic receipts.

✓ Processor Path Eliminated: Cleartext never leaves local RAM; GDPR Art. 28 processor obligations eliminated under Art. 32 safe harbor.

✓ Statutory Legal Shield: DPO & CISO receive provable due diligence evidence, eliminating personal administrative liability.

Result: Rapid InfoSec & Legal Sign-Off Immediate Revenue

Accelerate your B2B enterprise sales pipeline with the industry standard open trust seal.

Apply for Product Accreditation → Inspect A4 PDF Certificate →

ZERO-EGRESS SURFACES & PROTOCOL SOLUTIONS

5 Ways Developers Deploy ZTDS In-RAM Sanitization

From local IDE agent guardrails (Cursor, Claude Desktop) to RAG vector database de-identification: deploy vendor-neutral in-memory protection across every layer of your AI stack with 0.00 bytes cloud leakage.

SURFACE 01 · MODEL CONTEXT PROTOCOL Anthropic Registry: io.github.moxno/ztds

Stop Leaking Cleartext Secrets & Customer Records from Cursor and Claude Desktop

Developers frequently paste production API keys, database dumps, and sensitive customer records into AI coding composers. Traditional cloud DLP proxies cannot inspect desktop editor sessions without breaking TLS or injecting severe latency.

The open reference MCP server (ztds-mcp) executes locally as a child process over stdio JSON-RPC pipes. It sanitizes prompts, diffs, and tool arguments in volatile RAM in < 0.5ms before network serialization. Exactly 0.00 bytes socket egress.

✓ Zero Network Telemetry: Runs 100% on stdio pipes; opens zero outbound ports or sockets.

✓ Context-Preserving Surrogates: Replaces API keys and PII with tokens (e.g. [KEY_1], [EMAIL_1]) so AI retains full code logic.

✓ Reversible in Local RAM: AI completion responses are restored in process memory before rendering on your screen.

Full MCP Specification → mcpservers.org Profile → GitHub: ztds-mcp

Client Configuration

~/.cursor/mcp.json

{
  "mcpServers": {
    "ztds": {
      "command": "npx",
      "args": ["-y", "ztds-mcp"]
    }
  }
}

claude_desktop_config.json

{
  "mcpServers": {
    "ztds": {
      "command": "npx",
      "args": ["-y", "ztds-mcp"]
    }
  }
}

Terminal · 1-Line Setup

# Attach open ZTDS sanitization server to Claude Code CLI
claude mcp add ztds npx -y ztds-mcp

For 30 Specialized Enterprise Profiles (HIPAA, GLBA, PCI-DSS):

Replace ztds-mcp with certified commercial implementation @privacyscrubber/mcp-server.

SURFACE 02 · CI/CD AUDITOR Pre-Merge Static AST Linter

Fail the Build Before Raw PII Leaves Developer Machines

Once an engineer merges code containing unmasked LLM API calls, your organization incurs immediate GDPR Article 28 data processor liability. Remedying enterprise data leaks post-merge averages $4.88 million.

npx ztds-audit --strict inspects your abstract syntax tree (AST) in < 10ms. It verifies that every outbound AI connector is wrapped by an attested in-RAM sanitization boundary, producing deterministic cryptographic receipts.

✓ Zero API Keys Required: Performs 100% offline static analysis of code repositories.

✓ Ed25519 Receipts: Issues verifiable cryptographic proofs for compliance and audit logs.

✓ GitHub Actions Ready: Drop into CI/CD workflows to block regressions automatically.

Run Perimeter Scanner → GitHub Actions Template →

Terminal · Pre-Commit Verification

# Run automated invariant scan across repo
npx ztds-audit --strict

[PASS] Invariant 1: 0 unmasked egress sinks detected
[PASS] Invariant 2: Bijective token mapping verified
[PASS] Invariant 3: Zero-telemetry volatile RAM execution (<8ms)
[PASS] Invariant 4: Continuous subprocessor exclusion
[RECEIPT] Ed25519 SHA-256: 4f89b02a8f... VERIFIED CONFORMANT

SURFACE 03 · RAG VECTOR PIPELINES Pinecone, Qdrant, Chroma, LlamaIndex

When personal data or clinical health identifiers are converted into dense embedding vectors, fulfilling a GDPR Article 17 "Right to be Forgotten" deletion request requires unfeasible re-indexing of the entire high-dimensional vector space — costing weeks of cluster compute and thousands of dollars.

ZTDS sanitizes raw entities into deterministic surrogate tokens prior to embedding generation. Dense embeddings represent de-identified concepts with zero linkability to real identities, while the reversible token mapping stays exclusively in local volatile memory.

✓ Zero-Reindex Deletion: Discard local token map in RAM to render vector records permanently de-linked.

✓ Embedding Topology Preserved: Context-preserving surrogate tokens maintain vector semantic distance.

✓ LlamaIndex & LangChain Ready: Standard drop-in postprocessors and callbacks.

View RAG Blueprints → CISO Compliance Memo →

rag_ingestion.py · Invariant 1 + 2

from ztds import ZTDSCallbackHandler
from langchain_openai import ChatOpenAI

# 1. Attach client-side in-memory zero-trust hook
ztds = ZTDSCallbackHandler()

# 2. Outbound text is sanitized before vector embedding
llm = ChatOpenAI(callbacks=[ztds])
result = llm.invoke("Patient Alice Smith SSN 000-12-3456 records...")

# Emits: "Patient [NAME_1] SSN [SSN_1] records..." (0.00 B leak)
# Inbound: Restored in volatile RAM before returning

SURFACE 04 · HEADLESS SDK Node.js, TypeScript, Python, WASM

Sub-Millisecond In-Process Sanitization for Microservices & Agents

Cloud DLP proxies inject 250ms to 400ms of network overhead into every AI API call. In contrast, in-process volatile memory sanitization executes in < 0.8ms, operating entirely on host RAM with zero network hops.

Use the vendor-neutral open reference package @ztds/core (Apache 2.0) or certified commercial implementation @privacyscrubber/sdk for production multi-tenant environments.

✓ Zero Subprocessors: Completely eliminates third-party DPA obligations under GDPR Recital 26.

✓ High-Throughput: Handles 10,000+ operations/sec per CPU core in WebAssembly.

Explore Full SDK Docs → Browse Verified Registry →

sanitize.ts · @ztds/core

import { ZTDSClient } from '@ztds/core';

const ztds = new ZTDSClient();

// 1. Strip sensitive entities in local RAM
const { safePrompt, tokenMap } = ztds.sanitize(userInput);

// 2. Send de-identified tokens to external LLM
const response = await ai.generate(safePrompt);

// 3. Restore cleartext response in volatile memory
const clearOutput = ztds.restore(response, tokenMap);

SURFACE 05 · ENCLAVES & PROXIES AWS Nitro, Envoy WASM, SCIF

Hardware-Attested Isolation for Regulated Enterprise Environments

Regulated banking, aerospace, and healthcare institutions cannot rely exclusively on application-layer software isolation. Regulators demand verifiable cryptographic attestation.

ZTDS compiles to standalone WebAssembly modules and Nitro Enclave binaries. By deploying inside Envoy / Istio service meshes, unmasked outbound traffic is halted at the egress proxy before leaving the VPC perimeter.

✓ Cryptographic Attestation: TPM and Nitro attestation documents prove enclave integrity.

✓ Air-Gapped Licensing: Operates 100% offline via Ed25519 cryptographic license tokens.

Read RFC Invariants → OSF Latency Paper →

envoy-filter.yaml · Proxy-WASM

# Envoy Service Mesh Egress Filter
name: envoy.filters.http.wasm
typed_config:
  "@type": type.googleapis.com/envoy.extensions.filters.http.wasm.v3.Wasm
  config:
    name: "ztds_sanitizer"
    root_id: "ztds_in_ram_root"
    vm_config:
      runtime: "envoy.wasm.runtime.v8"
      code:
        local:
          filename: "/etc/envoy/ztds_sanitizer.wasm"

Architectural Paradigm Shift

The ZTDS Method vs. Legacy Cloud DLP

Legacy Cloud DLP = You hand keys to a third party. • ZTDS = You never let raw data leave the room.

Traditional enterprise security relies on intermediate cloud proxies that intercept cleartext traffic, introducing latency and adding third-party subprocessor liability under GDPR Article 28. The ZTDS Method moves the security boundary into volatile process memory: raw PII never crosses the network socket.

Legacy Approach

Cloud DLP & Proxy Gateways

Subprocessor Chain

Data Egress Path Unencrypted Socket Hop

User App → Cloud Proxy (+400ms) → LLM API

Network Latency Overhead: Adds 350ms to 600ms of external round-trip network lag to every LLM generation token stream.

Mandatory Subprocessor Liability: Cleartext leaves the device and enters a third-party SaaS proxy. Under GDPR Art. 28, this creates a new data subprocessor requiring full DPA & BAA legal agreements.

Centralized Cloud Honeypot: Intercepted customer prompts and decrypted payloads sit in proxy memory, disk logs, and cloud buffers, creating an attractive target for credential compromise.

Destructive Masking: Blind redaction with [REDACTED] breaks grammatical syntax, destroys LLM attention weight matrices, and produces hallucinatory responses.

The ZTDS Method (RFC v1.0)

In-RAM Client Sanitization

Safe Harbor Enclave

Memory Boundary Path 0.00 B Raw Egress

User App RAM ↔ ZTDS Enclave (<0.8ms) → LLM API (Tokens Only)

✓

Sub-Millisecond Execution: Executes in < 0.8ms in local WebAssembly or volatile Node.js runtime. Zero external network hops.

✓

GDPR Art. 28 Vendor Exemption: No personal data crosses the client boundary to the software vendor. Operating purely as a local computational utility, the architecture is fully exempt from GDPR Article 28 DPA requirements.

✓

Volatile Auto-Destruction: The ephemeral mapping table exists strictly in volatile RAM. It is never persisted to disk, database, or browser storage, and auto-destructs upon session termination.

✓

Bijective Lossless Tokens: Replaces entities with context-preserving surrogate tokens ([PATIENT_1], [IBAN_1]), preserving LLM reasoning and grammatical concordance.

RFC v1.0 Executive Abstract

3-Alternative AI Sanitization Architecture Matrix

IETF draft-sibiryakov-ztds-protocol-01

Zero-Trust Data Sanitization (ZTDS™ RFC v1.0) is an open cryptographic protocol that tokenizes sensitive PII, PHI, and secrets in local volatile RAM (< 0.8 ms latency, 0.00 B raw network egress) before any payload reaches an external LLM API or cloud socket—establishing technical safe harbor under GDPR Art. 32 and Israel Privacy Law Amendment 13 Sec. 17 without third-party subprocessor exposure.

Architecture Class	Execution Latency	Raw Cleartext Egress	Regulatory Posture (GDPR / Am. 13)	Token Semantics
1. ZTDS™ In-RAM Protocol (RFC v1.0)	< 0.8 ms (Volatile RAM)	0.00 B (Zero Network Egress)	Technical Safe Harbor (Art. 32 / Sec. 17) • Art. 28 Exempt	Bijective Surrogates ([PATIENT_1])
2. Cloud DLP API Proxy Gateways	250 – 600 ms (WAN Round-Trip)	100% Cleartext to 3rd-Party Proxy	Mandatory Art. 28 Subprocessor DPA & BAA Chain	Destructive Masking ([REDACTED])
3. Containerized NLP (Docker / Python)	35 – 90 ms (Container RPC)	Internal VPC Socket Hop	500MB+ Model Weights & Heavy DevOps Maintenance	One-Way Entity Replacement

LIVE IN-RAM DAG ENGINE

Test the ZTDS Method in Your Browser

Executes 100% locally in volatile browser memory. Open Network DevTools to verify 0.00 B network egress.

Sample Payload:

1. Raw Ingestion (Cleartext in RAM) 192 chars

2. Sanitized Egress (Sent to LLM) < 0.35 ms

Egress: 0.00 B

✓ Invariant 1: ΔEgress ≡ 0.00 B

✓ Invariant 2: T = M(V, C) Bijective

✓ Invariant 3: Volatile RAM Isolation

✓ Invariant 4: DPA Subprocessor Exclusion

30-Second Empirical Proof Zero Trust

Prove Zero Egress in Airplane Mode

Zero Trust means you never need to trust a vendor's promise. Disconnect your Wi-Fi right now, paste confidential customer records into the sandbox above, and run the sanitizer. Open your browser DevTools Network tab: exactly 0.00 bytes leave your machine.

WAN Packets: 0.00 B Cloud Sinks: 0

Transformer Attention Sub-2ms DX

Surrogates vs. Broken Redaction

Traditional redaction (like [REDACTED] or asterisks) flattens entropy and breaks LLM multi-head attention weights, causing hallucinations. ZTDS bracketed surrogates ([PERSON_1], [IBAN_2]) preserve semantic graph relationships, restored losslessly on your screen upon response.

Entropy Leakage: 0.00 bits Accuracy Loss: 0.00%

Enterprise Fast-Track Safe Harbor

Bypass 6-Week Security Reviews

Enterprise buyers stall on third-party subprocessor risk. Hand your CISO or customer procurement team our 1-page Safe Harbor Legal Memo: since cleartext never leaves device RAM, vendors eliminate GDPR Article 28 processor obligations and establish an Article 32 technical safe harbor.

Read CISO Memo → Download PDF

System Architecture & Invariants

The 4 Foundational Invariants of RFC v1.0

ZTDS is an open architectural specification and verification standard defined by four inviolable system boundaries enforced at the process memory perimeter and verified through automated network socket inspection.

01 Zero Egress

Zero External Egress Prior to Sanitization

ΔEgress ≡ 0.00 B

No personal data (PII, PHI, commercial secrets) may leave the local boundary in unmasked form. Enforced via socket byte-counting and audited via npx ztds-audit.

Rule: Network Boundary Lock

02 Bijective Tokens

Deterministic Reversible Tokenization

T = M(V, C) ↔ V = M‾¹(T)

Surrogate tokens preserve syntactic structure, grammar, and semantic roles for LLMs. The reverse mapping table is held strictly in volatile RAM for localized detokenization.

Rule: Context Concordance

03 In-RAM Isolation

Verifiable Memory Isolation

DiskWrite = 0 · DB = ∅

Ephemeral mapping tables cannot be written to disk, databases, localStorage, or third-party telemetry sinks. They auto-destruct when the execution session ends.

Rule: Zero Persistence

04 Safe Harbor

Complete Subprocessor Exclusion

GDPR Art. 32 & Art. 28 — Safe Harbor & Exclusion Path

Software functions purely as a local computing utility. Under GDPR Art. 32 and EDPB doctrine, zero vendor processing of PII eliminates the subprocessor chain under Art. 28 and establishes a technical safe harbor.

Rule: Statutory Demarcation

Read the Complete ZTDS RFC v1.0 Formal Specification →

Consortium & Ecosystem Accreditation

Rallying the Ecosystem Around the ZTDS Standard

ZTDS.ai is an open consortium, not a commercial SaaS vendor. We unite AI builders, enterprise CISOs, and security researchers to make client-side zero-trust sanitization the global industry norm.

TRACK A · BUILDERS Open Source & SaaS

Verified AI Applications

Sell to enterprise buyers without 90-day legal reviews. Prove 0.00 bytes cloud leakage with automated CLI verification, an official verified trust badge, and canonical registry inclusion.

Accreditation Benefits:

✓ Run automated CLI audit: npx ztds-audit

✓ Embed official ZTDS Verified™ dynamic SVG badge

✓ Listed in canonical /registry/ and llms.txt corpus

Verify Your Application → Browse Verified Registry →

TRACK B · ENTERPRISE CISOs & Legal

Corporate Adopters

Deploy internal LLMs and AI agents across your organization without signing new data processor agreements (DPAs) or expanding vendor attack surfaces.

Accreditation Benefits:

✓ Standardize internal AI usage on ZTDS Invariants

✓ 1-Page Safe Harbor & Legal Memorandum for Board

✓ Listed in Corporate Adopter pool (Founding: BrandMeWeb)

View Corporate Members → Download Legal CISO Pack →

TRACK C · ACADEMIA Researchers & Fellows

Fellows Council

Advance the mathematical and legal foundations of client-side privacy. Author peer-reviewed DOIs across Zenodo, OSF, SSRN, and IEEE-indexed publications.

Accreditation Benefits:

✓ Peer-reviewed DOIs on Zenodo, OSF, and SSRN

✓ Co-authorship on RFC v1.1 working group drafts

✓ Profile in the Global Fellows Council directory

Explore Fellows Council → Read Academic Treatise →

ZTDS.ai is governed as a vendor-neutral standard under Apache 2.0 & CC BY 4.0. Architecture sponsorship by BrandMeWeb. Reference engine implementation by PrivacyScrubber.

Read Governance Charter →

Reference Implementations

Deploy the Method in 4 Lines of Code

Deploy using the vendor-neutral open reference package @ztds/core (Apache 2.0) or the enterprise production engine @privacyscrubber/sdk. Both execute 100% locally in WebAssembly or host RAM without external API keys, telemetry, or network calls.

✓ Zero external configuration — operates completely in-RAM

✓ Sub-millisecond latency (< 0.8ms average execution)

✓ Connectors for LangChain, LlamaIndex, OpenAI, Anthropic, and MCP

Full SDK Documentation → Perimeter Network Scanner

// 1. Install open engine: npm i @ztds/core
import { ZTDSClient } from '@ztds/core';

// 2. Initialize in-memory zero-trust enclave
const ztds = new ZTDSClient();

// 3. Sanitize prompt in volatile RAM prior to external egress
const { safePrompt, tokenMap } = ztds.sanitize(rawUserPrompt);

// 4. Transmit safe surrogate tokens to external cloud LLM (0.00 B PII leak)
const aiResponse = await openai.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: safePrompt }]
});

// 5. Reversibly detokenize response locally in volatile RAM
const cleanResult = ztds.restore(aiResponse.content, tokenMap);

// .cursor/mcp.json or claude_desktop_config.json
"mcpServers": {
  "ztds": {
    "command": "npx",
    "args": ["-y", "ztds-mcp"]
  }
}

// Claude Code CLI (1-line attach):
claude mcp add ztds npx -y ztds-mcp

# Audit codebase AST for Invariants 1-4 in <10ms:
npx ztds-audit --strict

[PASS] Invariant 1: 0 unmasked egress sinks
[PASS] Invariant 2: Bijective token mapping verified
[PASS] Invariant 3: Zero-telemetry volatile RAM execution
[PASS] Invariant 4: Continuous subprocessor exclusion
[RECEIPT] Ed25519 SHA-256: 4f89b02a8f... VERIFIED

from ztds import ZTDSCallbackHandler
from langchain_openai import ChatOpenAI

# Prevents GDPR Art. 17 deletion liability in vector DBs
ztds = ZTDSCallbackHandler()
llm = ChatOpenAI(callbacks=[ztds])

# Prompts & vectors receive surrogate tokens ([NAME_1], [IBAN_1])
response = llm.invoke("Customer Jane Doe wire IBAN DE89370400440532013000")
# In-memory restoration ensures 0.00 B cleartext network leak

LEGAL & COMPLIANCE MEMORANDUM

Why ZTDS-Compliant Architectures Do Not Require a DPA

Under GDPR Article 28 and EDPB doctrine, data processing obligations apply when personal data is processed by an external third party. Because the ZTDS Method executes 100% in client RAM with 0.00 bytes transmitted to vendor servers, the vendor never qualifies as a "data processor", establishing an Article 32 technical safe harbor.

✓ GDPR Art. 32 Safe Harbor & Art. 28 Elimination

✓ HIPAA Safe Harbor 18 Exclusion

✓ EU AI Act High-Risk Shield

✓ SIG Lite & CAIQ Ready Controls

CISO Procurement Pack

Institutional Memorandum & Security Pack

Hand this directly to your enterprise buyers, legal counsel, and data protection officers.

Download 1-Page CISO Memo (PDF) → Download Complete Security Pack (ZIP) Open CISO Risk & ROI Calculator →

TECHNICAL ARCHITECTURE & COMPLIANCE

Frequently Asked Questions

Answers for developers, systems architects, and enterprise compliance officers.

How does ZTDS protect sensitive data in Model Context Protocol (MCP) for Cursor and Claude Desktop? ↓

The open reference MCP server (ztds-mcp) executes locally on your workstation as a child process over stdio JSON-RPC pipes. It intercepts prompts, code diffs, and tool arguments in volatile RAM before network transmission.

Cleartext API keys, internal IPs, and customer PII are replaced with contextual surrogate tokens ([KEY_1], [EMAIL_1]) so AI reasoning is unaffected, while zero bytes of sensitive data leave the machine. The mapping table is preserved strictly in client volatile memory and re-applied locally to responses before display.

How does ZTDS eliminate data processor liability under GDPR Article 28? ↓

Under GDPR Article 28, a Data Processing Agreement (DPA) is required when external third parties process personal data on your behalf. Because the ZTDS protocol executes 100% inside client volatile host RAM with 0.00 bytes of raw personal data transmitted over the network, ZTDS software never receives personal data and is physically eliminated from the data processor supply chain.

By neutralizing direct PII in local RAM and reducing entropy across vertical profiles, ZTDS establishes a proactive technical safe harbor under GDPR Article 32 (Technical & Organizational Measures) and Israel Protection of Privacy Law Amendment 13 Section 17, drastically mitigating statutory liability and eliminating subprocessor risk assessments.

What is the difference between open reference ztds-mcp and @privacyscrubber/mcp-server? ↓

ztds-mcp is the open-source reference implementation published under Apache 2.0 by the ZTDS Consortium, providing baseline universal entity sanitization (emails, keys, credit cards, SSNs, IPs) for individual developers.

@privacyscrubber/mcp-server is the certified enterprise production engine engineered by BrandMeWeb, including 30 specialized industry profiles (HIPAA clinical PHI, banking routing/IBANs, legal work product, cloud IAM credentials) and offline Ed25519 cryptographic token licensing for corporate teams.

How can I independently verify zero external network egress in my own environment? ↓

ZTDS conforms to the Airplane Mode Verification Protocol: load the web playground or run the CLI auditor npx ztds-audit --strict, physically disconnect your network interface (or enable Airplane Mode), and observe that all sanitization, detokenization, and verification execute with 100% functionality without error.

You can also run Wireshark or inspect the Chrome DevTools Network panel to confirm that exactly 0 network requests are dispatched during processing.

ZTDS^™ (Zero-Trust Data Sanitization) The Open Standard & Enterprise AI Trust Seal

How ZTDS Certification Accelerates B2B Enterprise AI Sales

Unlock Enterprise B2B Sales · Turn InfoSec Objections into Instant Sign-offs

Statutory Risk Minimization & Safe Harbor · Zero Personal Officer Liability

The Enterprise AI Adoption Chasm: 6 Months vs. 48 Hours

5 Ways Developers Deploy ZTDS In-RAM Sanitization

Stop Leaking Cleartext Secrets & Customer Records from Cursor and Claude Desktop

Fail the Build Before Raw PII Leaves Developer Machines

Sub-Millisecond In-Process Sanitization for Microservices & Agents

Hardware-Attested Isolation for Regulated Enterprise Environments

The ZTDS Method vs. Legacy Cloud DLP

Cloud DLP & Proxy Gateways

In-RAM Client Sanitization

3-Alternative AI Sanitization Architecture Matrix

Test the ZTDS Method in Your Browser

Prove Zero Egress in Airplane Mode

Surrogates vs. Broken Redaction

Bypass 6-Week Security Reviews

The 4 Foundational Invariants of RFC v1.0

Zero External Egress Prior to Sanitization

Deterministic Reversible Tokenization

Verifiable Memory Isolation

Complete Subprocessor Exclusion

Rallying the Ecosystem Around the ZTDS Standard

Verified AI Applications

Corporate Adopters

Fellows Council

Deploy the Method in 4 Lines of Code

Why ZTDS-Compliant Architectures Do Not Require a DPA

Frequently Asked Questions

Download Official CISO Compliance Memorandum

Official Memorandum Generated

ZTDS™ (Zero-Trust Data Sanitization) The Open Standard & Enterprise AI Trust Seal

How ZTDS Certification Accelerates B2B Enterprise AI Sales

Unlock Enterprise B2B Sales · Turn InfoSec Objections into Instant Sign-offs

Statutory Risk Minimization & Safe Harbor · Zero Personal Officer Liability

The Enterprise AI Adoption Chasm: 6 Months vs. 48 Hours

5 Ways Developers Deploy ZTDS In-RAM Sanitization

Stop Leaking Cleartext Secrets & Customer Records from Cursor and Claude Desktop

Fail the Build Before Raw PII Leaves Developer Machines

Immunize Vector Databases Against the GDPR Article 17 Deletion Trap

Sub-Millisecond In-Process Sanitization for Microservices & Agents

Hardware-Attested Isolation for Regulated Enterprise Environments

The ZTDS Method vs. Legacy Cloud DLP

Cloud DLP & Proxy Gateways

In-RAM Client Sanitization

3-Alternative AI Sanitization Architecture Matrix

Test the ZTDS Method in Your Browser

Prove Zero Egress in Airplane Mode

Surrogates vs. Broken Redaction

Bypass 6-Week Security Reviews

The 4 Foundational Invariants of RFC v1.0

Zero External Egress Prior to Sanitization

Deterministic Reversible Tokenization

Verifiable Memory Isolation

Complete Subprocessor Exclusion

Rallying the Ecosystem Around the ZTDS Standard

Verified AI Applications

Corporate Adopters

Fellows Council

Deploy the Method in 4 Lines of Code

Why ZTDS-Compliant Architectures Do Not Require a DPA

Frequently Asked Questions

ZTDS^™ (Zero-Trust Data Sanitization) The Open Standard & Enterprise AI Trust Seal