The Engine
Admin authentication required
Or use an existing operator token from the Operator Console

Handvantage The Engine — Infrastructure Control Plane

Unified control and monitoring for zero-trust infrastructure

The Maker
Status Active
Uptime —
Request Count —
Open Maker Console →
The Checker
Status Active
Last Grade —
Last Assessment —
Open Checker Dashboard →
The Engine
Status Active
Policies Loaded —
Enforcement deny-by-default
Local Memory
Status Active
Conversation SQLite + FTS5
Audit Trail Immutable JSONL
LLM Cost Control-plane: $0 · Desk LLM: your provider (BYOK/on-box)
Compliance Evidence Report
Generate a branded PDF with live posture grade, policy inventory, audit summary, and compliance control matrix. Ready to forward to a CISO.

Platform Architecture

AI Operations
Core
Business Operator (8089)
Engine / Admin (8090)
Security Operator (8091)
Checker (internal)
Hindsight Memory (8765)
Identity & Access
Internal
Keycloak SSO (8180)
PostgreSQL (5432)
Workspace Services
Internal
Nextcloud (Files/Cal)
OnlyOffice (Docs)
Synapse (Chat)
Element Web (UI)
Communication
Internal
Stalwart (Email)
Jitsi (Video)
LLM Layer
Internal
Ollama (Host LLM)
NemoClaw (Sandbox)
Edge / Proxy
External
Caddy (reverse proxy)
Cloudflare Tunnel
Audit & Memory Architecture
Active — Zero LLM Cost
Audit Trail
✓
immutable JSONL logs
Conversations
✓
SQLite per-operator
Full-Text Search
✓
FTS5 indexed
Checker Findings
✓
local assessment logs
Policy Engine
✓
runtime enforcement
Control-Plane LLM
$0
Desk LLM: your provider (BYOK/on-box)
—
Services Running
—
Policies Loaded
—
Policy Self-Grade
—
Audit Events
—
Uptime
Enforcement: deny-by-default
Loading policies...

Live Audit Event Stream

TimeTypeSourceMessage
Loading...
Environment
Loading environment...
Service Architecture
Loading services...
Deployment Commands
Stop All Services
View Logs
Rebuild Single Service
—
Policy Self-Assessment Grade
Self-assessment from policy configuration
Independent Checker: checking status…

Framework Compliance Scores

Loading compliance data...

Assessment Details

Grade is calculated in real-time from loaded policy YAML files mapped against 11 compliance frameworks: NIST AI RMF, ISO 42001, EU AI Act, SOC 2, PCI DSS v4.0, HIPAA, FINRA, FedRAMP, PIPEDA, Privacy Act (Canada), and TBSDADM (Treasury Board Directive on Automated Decision-Making). When the independent Checker is online, its grade is shown for cross-validation.
Enforcement mode: deny-by-default
Grade formula: Average of per-framework control coverage → letter grade
A: ≥93% B: ≥83% C: ≥73% D: ≥60% F: <60%

Framework Requirement Mapping

Framework Control ID Requirement Test Coverage
Loading framework mappings...
10/10
OWASP Agentic Categories — Controls Mapped
OWASP Top 10 for Agentic Applications (2026) — control coverage self-assessed from the bundled test catalog. Design-time mapping, not an independent live validation.
Source: bundled OWASP Agentic control catalog

7-Layer Defense Architecture

L7
Supply Chain — SHA-256 manifest, HMAC signing, provenance
L6
Inter-Service Auth — JWT tokens, HMAC payloads, replay protection
L5
Trust Boundaries — 8 zones, zone auth, confidence scoring, exfil detection
L4
Memory Safety — injection scan, PII redact, integrity hash, metadata strip
L3
Tool Guardrails — rate limits, param validation, call budgets, confirmation gates
L2
Prompt Defense — NemoClaw-Guard regex, LLM classifier, canary tokens, base64 decode
L1
Policy Engine — GATEWAY_ROLE isolation, Maker/Checker, plan validation, audit trail

Coverage Matrix — click to expand

▶ ASI-01 Goal Hijacking & Prompt Injection ✓ Covered 47 tests S63
Attackers inject malicious instructions into prompts to hijack agent goals, exfiltrate data, or bypass safety controls.
Controls
  • NemoClaw-Guard regex engine (5 pattern families)
  • LLM-based semantic classifier
  • Canary token detection & alerting
  • Base64/encoding decode before scan
  • Multi-turn drift detection
Framework Cross-References
NIST AI RMF GV-1.3 ISO 42001 A.7.2 EU AI Act Art. 15 Zero Trust: Verify Explicitly
▶ ASI-02 Tool Misuse & Abuse ✓ Covered 26 tests S64
Agents invoke tools with unsafe parameters, excessive frequency, or beyond intended scope, causing unintended side effects.
Controls
  • ToolGuardrails: per-tool rate limiter (sliding window)
  • Parameter validation with type checking
  • Call budget enforcement per conversation
  • Confirmation gates for destructive operations
Framework Cross-References
NIST AI RMF MP-4.1 ISO 42001 A.8.4 EU AI Act Art. 14 Zero Trust: Least Privilege
▶ ASI-03 Excessive Agency & Privilege ✓ Covered 6 tests S63
Agents granted more permissions than needed, enabling lateral movement or unauthorized access to sensitive systems.
Controls
  • GATEWAY_ROLE isolation (Business/Security/Engine)
  • TOOLS_BY_ROLE per-container tool allowlists
  • 3-container architecture separation
  • JWT-authenticated cross-service calls
Framework Cross-References
NIST AI RMF GV-3.2 ISO 42001 A.6.2 EU AI Act Art. 9 Zero Trust: Least Privilege
▶ ASI-04 Supply Chain Vulnerabilities ✓ Covered 22 tests S65
Compromised tools, poisoned models, or tampered configurations introduced through the supply chain.
Controls
  • SupplyChainVerifier: SHA-256 file hashing
  • HMAC-signed manifest verification
  • Tool provenance tracking
  • 3 verification modes (strict/warn/audit)
Framework Cross-References
NIST AI RMF MP-2.3 SLSA Level 3 ISO 42001 A.10.2 EU AI Act Art. 15
▶ ASI-05 Unsafe Code Execution ✓ Covered 6 tests S63
Agents execute arbitrary code, shell commands, or scripts without proper sandboxing, enabling system compromise.
Controls
  • NemoClaw sandbox environment
  • No inline exec/eval in gateway
  • Shell injection regex detection
  • Execution timeout enforcement
Framework Cross-References
NIST AI RMF MP-4.1 ISO 42001 A.8.2 Zero Trust: Assume Breach
▶ ASI-06 Context Poisoning ✓ Covered 14 tests S64
Malicious content injected into agent memory or context windows, corrupting future decisions and outputs.
Controls
  • MemorySanitizer: injection pattern detection
  • PII redaction before storage
  • SHA-256 integrity hashing on recall
  • Metadata stripping
  • Unicode normalization
Framework Cross-References
NIST AI RMF MG-3.2 ISO 42001 A.7.4 EU AI Act Art. 10 Zero Trust: Verify Explicitly
▶ ASI-07 Inter-Agent Communication ✓ Covered 23 tests S65
Unauthenticated or unsigned messages between agents enabling spoofing, replay attacks, or privilege escalation.
Controls
  • InterServiceAuth: JWT token generation & verification
  • HMAC payload signing
  • Replay protection with nonce cache
  • Audit trail for all cross-service calls
Framework Cross-References
NIST AI RMF GV-6.1 ISO 42001 A.9.3 SLSA: Build Integrity Zero Trust: Verify Explicitly
▶ ASI-08 Cascade Failures ✓ Covered ✓ S63
A failure in one agent or tool cascades across the system, causing widespread outage or data corruption.
Controls
  • Retain queue for Hindsight resilience
  • Multi-level timeout enforcement
  • Budget caps per conversation
  • Concurrency semaphore limiting
  • Health check endpoints with auto-recovery
Framework Cross-References
NIST AI RMF MG-2.4 ISO 42001 A.5.4 Zero Trust: Assume Breach
▶ ASI-09 Trust Exploitation ✓ Covered 14 tests S64
Agents blindly trust inputs from other agents or external sources, enabling privilege escalation and data exfiltration.
Controls
  • TrustBoundaryEnforcer: 8 trust zones
  • Per-zone authorization checks
  • Confidence scoring for cross-boundary requests
  • Exfiltration detection patterns
Framework Cross-References
NIST AI RMF GV-1.6 ISO 42001 A.6.4 EU AI Act Art. 9 Zero Trust: Never Trust
▶ ASI-10 Rogue Agents & Autonomy ✓ Covered 10 tests S63
Agents operate beyond their intended scope, spawn unauthorized sub-agents, or accumulate unchecked autonomy.
Controls
  • PlanValidator: subtask limits & cycle detection
  • Budget enforcement per plan
  • Maker/Checker dual-operator architecture
  • Comprehensive audit logging
Framework Cross-References
NIST AI RMF GV-3.2 ISO 42001 A.4.3 EU AI Act Art. 14 Zero Trust: Least Privilege

Framework Coverage Summary

7/7
NIST AI RMF Controls
8/8
ISO 42001 Annex A
5/5
EU AI Act Articles
2/2
SLSA / Zero Trust

Hardening Roadmap

S63 ✓ Document + test existing controls (ASI-01, 03, 05, 08, 10) S64 ✓ Targeted hardening: tool rate limits (ASI-02), memory sanitization (ASI-06), trust boundary (ASI-09) S65 ✓ Advanced controls: supply chain verification (ASI-04), inter-service JWT auth (ASI-07)

Automated Test Results

ASI-01: Goal Hijacking
Role hijack: 10 tests
System prompt exfil: 6 tests
DAN/Jailbreak: 8 tests
Delimiter spoof: 5 tests
Shell execution: 3 tests
Unicode bypass: 3 tests
Base64 obfuscation: 1 test
Canary tokens: 2 tests
Indirect injection: 3 tests
Multi-turn drift: 2 tests
Secret exfil: 4 tests
47/47 passed (100%)
ASI-03/05/10: Structural Controls
Tool role isolation: 3 tests
Write confirmation: 1 test
Gateway role source: 2 tests
No inline exec/eval: 1 test
Code exec delegation: 2 tests
Shell injection regex: 3 tests
Subtask limits: 3 tests
Circular dep detection: 3 tests
Prompt length limits: 2 tests
Budget enforcement: 1 test
Dep bounds: 1 test
22/22 passed (100%)
ASI-02/06/09: S64 Hardening
Rate limiter: 7 tests
Param validation: 8 tests
Call budget: 8 tests
Guardrails integration: 3 tests
Injection detection: 6 tests
PII redaction: 3 tests
Integrity hash: 2 tests
Metadata strip: 1 test
Unicode normalization: 1 test
Zone authorization: 6 tests
Cross-boundary: 3 tests
Confidence scoring: 5 tests
54/54 passed (100%)
ASI-04/07: S65 Supply Chain + Comms
File hashing: 4 tests
Manifest sign/verify: 5 tests
Supply chain verifier: 9 tests
Single file verify: 1 test
Tool tamper detection: 1 test
Summary format: 1 test
Token generation: 4 tests
Token verification: 8 tests
Payload signing: 5 tests
Nonce cache: 5 tests
Auth headers: 3 tests
Audit trail: 2 tests
45/45 passed (100%)
Hindsight Memory
Checking status...
Loading memory banks...

Search Across All Banks

Reflect (AI Reasoning)

Enterprise Hindsight Architecture
Per-User Banks: Each operator gets isolated memory (user-{id})
Department Banks: Shared context within business units (dept-{name})
Org Decisions: Cross-team decisions, rulings, exceptions auto-captured
Policy Knowledge: Governance policies indexed for LLM grounding
Audit Banks: Searchable supplement to immutable JSONL audit trail
Entity Extraction: Configurable (local LLM or cloud provider)

Platform Users

Create New User

Reset Password

User:

Change Role

User:  ·  Current:
HR data (roster, headcount, leave, analytics) is limited to admins and users granted here. Regular staff keep self-service only.
CRM records are limited to admins and users granted here. Without this grant the CRM tools decline every request, whatever the phrasing.
The Bookkeeper role grants access to the Month-End Close dashboard + AI Bookkeeper. For SSO users this is written to the identity provider so it persists across logins.
Username Display Name Role Department Status Last Login Actions
Loading users...