Features — WITZ Guardian

SP-03 · Core

Dynamic Risk Score API

Real-time risk assessment for any agent on Base Network. Four weighted factors produce a 0-100 score updated with every transaction.

On-chain history (40%): anomaly rate + unresolved alerts
Network behavior (20%): unknown destinations + tx frequency
Market conditions (25%): ETH volatility + volume changes
Macro context (15%): KYA status + tier + agent age

FreeProEnterprise

0-100

Risk Score updated per-transaction

SP-06 · AI Engine

LSTM Anomaly Detection

Per-agent LSTM Autoencoder with Self-Attention learns normal behavior patterns and detects deviations that rule engines cannot catch.

10-feature vector per transaction (temporal + behavioral)
Sequence analysis: 20-transaction sliding window
Auto-training when 100+ transactions accumulated
Shadow mode validation before going live
Backtested: F1=0.917, AUC-PR=0.895

ProEnterprise

91.7%

F1 Score across 10 agent profiles

SP-01 · Identity

KYA Protocol (Know Your Agent)

Verifiable on-chain identity for autonomous agents. NFT Soulbound on Base Network containing agent metadata, creator info, and audit links.

Non-transferable NFT Soulbound (AgentIdentity.sol)
Metadata URI with description, purpose, audit links
Status lifecycle: Unverified → Pending → Verified
Queryable by any smart contract on Base

FreeProEnterprise

NFT

Soulbound identity on Base Network

SP-02 · Trust

Portable Reputation Score

Cumulative trust score that grows with clean operation. Agents with score 70+ are publicly queryable, reducing protocol onboarding from 4 days to minutes.

5 factors: clean history, continuity, compliance, KYA, model quality
Inactivity decay: -2 pts/week (cap 30 pts)
On-chain via ReputationScore.sol
Shareable Passport URL for cross-protocol trust

FreeProEnterprise

Reputation Levels

0-29 · New

30-49 · Building

50-69 · Established

70-89 · Trusted (public)

90-100 · Elite

SP-04 · Containment

Circuit Breaker

One-click containment of compromised agents. Per-agent control pauses operations instantly with immutable on-chain evidence trail.

Per-agent containment (don't affect other agents)
Response time: under 10 seconds
On-chain audit log via AuditRegistry.sol
Manual release with confirmation

ProEnterprise

<10s

Time to contain a compromised agent

SP-05 · Emergency

War Room Mode

Dedicated emergency interface for incident management. Critical alerts only, circuit breaker controls, and incident timeline in one view.

Critical/High alert feed with auto-refresh
Circuit breaker panel for all agents
Impact estimator: value at risk
Escalation actions: Telegram + Discord + Email

Enterprise

Incident Flow

Alert received (Telegram)

War Room activated

XAI: identify root cause

Circuit Breaker engaged

On-chain evidence logged

SP-07 · Explainability

XAI — Explainable AI

Every anomaly detection comes with a human-readable explanation. Powered by LSTM attention weights, translated into plain language for non-technical stakeholders.

Feature deviation visualization (which behavior changed)
Attention heatmap (which transactions triggered it)
Natural language narratives for compliance reports
Export-ready for LP reporting (Enterprise)

Pro (basic)Enterprise (full)

Example XAI Output:

"TraderBot_v2 spent 3.2x more gas than usual interacting with a contract it had never used before, at 3:42 AM — outside its normal operating hours."

SP-08 · Phase 2 · Decentralized

Sentinel Network

Distributed observation network where independent operators contribute blockchain data and earn commissions proportional to verified value.

Low barrier: no GPU, no staking, no expertise required
Commissions in stablecoins (not speculative token)
More sentinels = better AI = better alerts = more clients
Open-source client (Phase 2+)

Phase 2

N×

More sentinels = exponentially better detection

Everything You Need to Govern AI Agents

Start Protecting Your Agents