Secure by Design

Enterprise security, engineered in

Fortress Pipeline: 8 stages, 9 AI guards, encrypted data, physically isolated tenants. Security engineered into the Agent OS, not bolted on.

Fortress Pipeline · 8 stagesOWASP LLM Top 10 & Agentic AI Threats defenseAES-256-GCM + HMAC blind indexCloud · physical tenant isolationSelf-hosted · secure Docker distributionRBAC access controlAgent knowledge access control
AGENT LAYER
Multi-layer AI security guards

Multi-layer AI security guards · OWASP Top 10 defenses · HITL takeover

DATA LAYER
PII encryption · search-while-encrypted · per-group keys

AES-256-GCM authenticated encryption · encrypted search · per-group cryptographic keys

INFRASTRUCTURE LAYER
Physical tenant isolation · encrypted container images

Dedicated VMs · separated DB tables · encrypted container images · signed supply chain

SECURITY FOUNDATIONS

Security across infrastructure, data, and agents

Infrastructure Layer

Physical tenant isolation

Dedicated VMs and separated database tables per tenant. Not row-level: infrastructure-boundary isolation.

Infrastructure Layer

Encrypted container images

Docker images encrypted layer-by-layer with per-customer keys. Supply chain signed, tamper-evident.

Data Layer

AES-256-GCM PII encryption

Sensitive fields are encrypted at the application layer with authenticated GCM mode. Keys rotated via KeyVault.

Data Layer

Search-while-encrypted index

Search encrypted fields by exact match without ever decrypting. Industry-standard secure indexing.

Data Layer

Per-group dedicated encryption keys

Independent cryptographic keys per group; compromise of one group cannot expose others. Encrypted key backup.

Agent Layer

Multi-layer AI security guards

Prompt injection, knowledge poisoning, and PII leakage defended at platform level: aligned with OWASP Top 10 for LLM.

AGENT ACCESS CONTROL

Knowledge access, layered by agent tier

Every agent is classified by access tier: and knowledge reach is scoped accordingly. A public concierge cannot see internal group data; a personal assistant cannot see other users' private notes.

Public scope
Public · Concierge
Group scope
Export · Group
Personal scope
Private · Assistant

Private · Assistant

Personal scope

Access only to the individual user's private knowledge, notes, and conversation history. Isolated from group and public knowledge unless explicitly granted.

Export · Group

Group scope

Access to shared group knowledge (team documents, collective memory). Private user data remains invisible. Bound to the group's dedicated encryption key.

Public · Concierge

Public scope

Customer-facing agents restricted to sanitized public knowledge. No internal data, no user history: safe for signage, websites, and external channels.

OWASP TOP 10 · LLM & AI AGENT THREATS

Defenses mapped to OWASP for AI and Agent threats

Every AICLUDE deployment ships with platform-level defenses aligned with OWASP Top 10 for LLM Applications and the emerging OWASP for AI Agent threat catalog: not bolted on, but built into the pipeline.

Prompt Injection
Multi-stage input guards · system prompt isolation
Sensitive Information Disclosure
AES-256-GCM PII encryption · output filters
Data & Knowledge Poisoning
Source provenance · graph-based verification
Supply Chain
Signed containers · encrypted images · SBOM
Excessive Agency
Tool permission scopes · Human-in-the-loop takeover
Agent Hijacking
Agent-tier access control (Private/Export/Public)
Tool Misuse
MCP permission model · audit logs
Unbounded Consumption
Token quotas · rate limits · tenant isolation
AICLUDE VS: SKILL & MCP SAFETY SCANNER

Every Skill and MCP server, scanned before you install

AICLUDE VS continuously pre-scans the global Skill and MCP ecosystem: thousands of packages from npm, GitHub, Smithery, and more. 7 parallel scan engines analyze each package across 10 AI-specific threat categories, so only vetted, safe Skills and MCP servers reach your agents.

7
Parallel Scan Engines
10
AI Threat Categories
Thousands
Packages Scanned Daily
  • 7 parallel scan engines (SAST · SCA · Tool Analyzer · DAST · Permission · Behavior · Malware)
  • 10 AI threat categories (Prompt Injection · Tool Poisoning · Command Injection · …)
  • Risk Score 0–100 with 5 severity levels · SBOM auto-generated
  • Scan Search · Risk Reports · Connected Agents · MCP / Skill integration
Explore AICLUDE VS
vs.aiclude.com/dashboard
Average Risk Score
LIVE
12/ 100LOW RISK
@modelcontextprotocol/server-filesystem
8
npm:agentic-skill-web-search
14
smithery:slack-mcp
42
github:acme/mcp-sql-tools
19
SAST
SCA
Tool
DAST
Perm
Behav
Mal
THREAT CATEGORIES DETECTED
Prompt InjectionTool PoisoningCommand InjectionCredential HarvestingData ExfiltrationSupply ChainPermission EscalationMalicious BehaviorObfuscated CodeBackdoors
DEPLOYMENT SECURITY

Same Docker images. Your choice of infrastructure.

Cloud: physical tenant VMs

Each Cloud tier from Pro up runs on a dedicated tenant VM. Starter uses a shared tenant pool with per-tenant resource quotas.

Self-hosted: encrypted binaries

Encrypted container images delivered to your infrastructure via private registry. License server check-in with 7-day offline grace.

Air-gapped: Enterprise

Full offline / air-gapped deployment available for Self-hosted Enterprise. Model ONNX packages + encrypted image bundle delivered physically.

Ready to see it in action?

Open AICLUDE VS to audit your agents, or contact us for a security deep-dive.

Open AICLUDE VS