ZTDS.ai Open AI Security Standard
Get Certified →
CLIENT-SIDE VOLATILE RAM ENVIRONMENT · ZERO SERVER EGRESS

Interactive Pipeline Inspector

Simulate real-time in-memory sanitization across all 4 RFC v1.0 pipeline stages: Egress Interception, Tokenization, Enclave Isolation, and Loopback De-tokenization.

Industry Profiles:
Stage 01 Quarantined
1. Egress Interception

Traps socket write before WAN dispatch (RFC v1.0 Invariant 1).

184 chars 0.00 B Egress
Stage 02 Deterministic
2. Ephemeral Tokenizer

1-to-1 reversible mapping; I(X; T) = 0.00 bits mutual information.

5 tokens 0 Collisions
Stage 03 Volatile RAM
3. Enclave Execution

WASM memory scope; M_map ∩ Storage = ∅; 0 disk telemetry.

<0.25 ms 0 Storage Keys
Stage 04 Zero-Subprocessor
4. Loopback Reveal

Client-side stream de-tokenization; GDPR Art. 28 DPA exemption.

100% Reversible Auto-Zeroize
Pipeline State: READY (RAM INITIALIZED)
1. Ingestion Buffer (Cleartext Quarantined in Client RAM) Live Input Buffer
Buffer Size: 184 B · WAN Sockets: 0.00 B Egress
Volatile Scoped
2. Sanitized Model Payload (Surrogate Tokens) Latency: 0.22ms
Reversible Token Rate: 100% · Collisions: 0.0000
Shannon I(X;T) = 0.00b

Volatile In-Memory Token Map (Ephemeral RAM Dump)

Isolated strictly to volatile execution memory. Conforming to Theorem 2: ∅ non-volatile disk persistence.

{ /* Initializing volatile RAM token map */ }
Theorem 2 Lifecycle Constraint: lim_{t → t_destroy} M_map = ∅ M_map ⊆ VolatileRAM
Invariant 1: Zero-Egress
PASS (0.00 B WAN)

Cleartext never touches network socket frames.

Invariant 2: Reversible
PASS (0 Collisions)

Deterministic surrogate placeholder restoration.

Invariant 3: RAM Isolated
PASS (WASM Enclave)

Zero disk, cookie, or local storage writes.

Invariant 4: Zero DPA
PASS (Art. 28 Exempt)

No third-party subprocessor chain of custody.

Developer SDK · Reference Implementation

Embed ZTDS Invariant 1 & 2 into Production RAG

Protect your LangChain, LlamaIndex, or agentic pipelines in 2 lines of code. Zero third-party network egress. Powered by reference engine PrivacyScrubber.

npm install @privacyscrubber/sdk
View 10 Architectural Blueprints → Node · Python · WASM
Consortium Working Group · WG-2

Contribute Domain Entity Rulesets

Help expand the open standard. Submit high-precision regex patterns, financial & IBAN checksum validators, or formal verification benchmarks via GitOps pull request.

git checkout -b rules/eu-iban-profile
Submit Rule via PR → Working Group 2 (WG-2)
Direct Answer Knowledge Base

Interactive Zero-Trust Pipeline Inspector FAQ

What does the ZTDS Pipeline Inspector simulate in the browser? ↓
The Pipeline Inspector executes a live four-stage Directed Acyclic Graph (DAG) inside browser volatile memory: (1) Raw Prompt Ingestion, (2) Deterministic In-RAM Surrogate Tokenization, (3) Simulated Cloud LLM Reasoning over synthetic tokens, and (4) Local Bijective Restoration.
Does the Pipeline Inspector transmit test prompts to any backend server or external LLM API? ↓
No. Every stage of the Pipeline Inspector runs 100% locally in your browser JavaScript heap with 0.00 bytes of network egress. You can verify this by opening browser DevTools Network inspection or enabling Airplane Mode while testing.
How does the Inspector demonstrate Theorem 2 volatile memory zeroization? ↓
During simulation, the Inspector displays the active ephemeral tokenMap holding surrogate-to-entity pairs in RAM. Clicking Flush Memory or completing the session immediately zeroizes all map entries, proving that zero residual PII persists in memory or storage.
How can developers integrate the inspected DAG pipeline into production Node.js or Python apps? ↓
Developers can copy the generated four-line integration blueprint using the open reference package (@ztds/core or ztds-python) or deploy the certified commercial engine (@privacyscrubber/sdk and @privacyscrubber/mcp-server) for 30 specialized industry profiles.