Mastrion GmbH
Console Login Meet Us →
INTERACTIVE SANDBOX & CLI

Live Sovereign Runtime Inspector

Test real-time dynamic PII redaction on arbitrary input text, inspect Radix Prefix Cache token reclaim proof, and run operational CLI commands.

1. Client Request Ingress

Select a scenario or type any arbitrary customer text, dates of birth, names, or accounts:

Type freely: Redacts in real time

2. Egress Pipeline Telemetry

Output produced by the Envoy / Istio ambient proxy before model inference:

RADIX PREFIX CACHE ENGINE

Prefix Cache Token Reclaim Proof (How 62% is Saved)

In enterprise deployments, applications repeatedly transmit identical system instructions, corporate policies, and schemas. Mastrion's Radix KV cache keeps these attention keys in GPU VRAM permanently, computing only the fresh prompt delta.

62.4%
Avg. Monthly Token Reclaim
Chunk 1: System Persona & Policy
100% IN-VRAM CACHE HIT
820 Tokens · Key Hash: 0x8f2a1b9c
"You are a compliant banking assistant bound by DORA and SEC rules. Always verify KYC status..."
GPU Prefill Time: 0.00ms (Reclaimed)
Chunk 2: Enterprise Schema & Context
100% IN-VRAM CACHE HIT
380 Tokens · Key Hash: 0x4d3e2f10
"Database definitions: accounts(id, iban, status), transactions(tx_id, amount, timestamp)..."
GPU Prefill Time: 0.00ms (Reclaimed)
Chunk 3: Unique User Query
FRESH ATTENTION COMPUTATION
38 Tokens · Input Delta
"Please summarize transaction for account..."
GPU Compute: 24.5ms (Delta Only)
Total Request Context: 1,238 tokens → 1,200 tokens (96.9%) served directly from local GPU VRAM
Time-to-First-Token: 1.82s → 0.12s (93.4% Latency Reduction)

Authentic `mastrionctl` CLI Sandbox

Run operational lifecycle commands formatted with real multi-line terminal outputs:

Quick Run:
operator@mastrion-gateway: ~ (zsh)
WAYPOINT PROXY ACTIVE