⛊ AEGIS OS v4.1

Autonomous AI Agent Security Platform — Built in Rust

88% of organizations reported AI agent security incidents. AEGIS OS is the first autonomous security platform that discovers, tests, and protects AI agents — automatically.

What is AEGIS OS?

AEGIS OS is an open-source autonomous security platform for AI agents. It wraps every agent in 12 security layers, discovers new agents on your network, tests them continuously, learns from attacks, and generates reports — all without human intervention.

cargo run -- autonomous     # Autonomous 24/7 security daemon
cargo run -- serve          # HTTP server on :8401
cargo run -- serve-tls      # HTTPS on :8443
cargo run -- red-team       # 16 attack vectors tested

Autonomous Mode — Works Alone

$ cargo run -- autonomous --cycles 3

[DAEMON] ═══ Cycle 1 ═══
[DAEMON] Phase 1: Discovery — Found 3 agents (OpenClaw, Goose, MCP)
[DAEMON] Phase 2: Testing — 15 tests × 3 agents = 45 tests
[DAEMON] Phase 3: Analysis — 1 threat: OpenClaw (9 CVEs)
[DAEMON] Phase 4: Blocking — OpenClaw auto-blocked
[DAEMON] Phase 5: Learning — 1 new rule generated
[DAEMON] Phase 6: Report — Daily report generated
[DAEMON] ═══ Cycle 1 Complete ═══

6-Phase Autonomous Cycle

Phase	Engine	What it does
1. Discover	Discovery Engine	Scans ports, MCP servers, Docker containers
2. Test	Continuous Testing	15 security tests per agent
3. Analyze	Threat Analysis	Risk scoring + known CVE matching
4. Block	Auto-Block	Blocks agents above risk threshold
5. Learn	Adaptive Learning	Creates new rules from attacks
6. Report	Autonomous Reporter	Daily/weekly/incident reports

32 Security Modules

Category	Modules
Core	CLI, Server (Axum), Orchestrator, Gateway
AI	9 backends: NIM, Groq, OpenAI, Gemini, Claude, Cosmos, Ollama, Mistral, DeepSeek
Security	Input Sanitization, Skill Vetting, Guardrails, Isolation, Credentials, Risk Scoring
AI Security	RAG Security Guard, Privacy Filter, Extraction Detector, Hallucination Detector, AI Watermark
Protocols	MCP Client/Server (20 tools), A2A Gateway, MCP-Kali (12 pentest tools)
Autonomous	Discovery Engine, Continuous Testing, Adaptive Learning, Autonomous Reporter, Autonomous Daemon
Infrastructure	Database, ROS2, Telemetry, Audit Logger

12 Security Layers

Layer 1:  API Gateway + TLS/HTTPS + RBAC (4 roles)
Layer 2:  AI Inference (9 backends + fallback chain)
Layer 3:  Policy Engine + Guardrails
Layer 4:  Input Sanitization (21 injection patterns)
Layer 5:  Skill Vetting Pipeline (static + deps + sandbox)
Layer 6:  Rate Limiting + DDoS Protection
Layer 7:  Tool Router + MCP Bridge (20 tools)
Layer 8:  MCP Server + A2A Gateway
Layer 9:  WASM Sandbox + Filesystem Jail
Layer 10: State Integrity Monitor (SHA256 + drift)
Layer 11: Credential Vault + Capability Tokens + TTL
Layer 12: Risk Scoring (4D) + Audit + SIEM Export

9 AI Backends

Provider	Model	Status
NVIDIA NIM	Llama 3.1 8B	✅ Verified
Groq	Llama 3.3 70B	✅ Verified
OpenAI	GPT-4o-mini	✅ Verified
Google Gemini	2.5 Flash Lite	✅ Verified
Anthropic Claude	Haiku 4.5	✅ Ready
NVIDIA Cosmos	Reason 7B	✅ Built
Ollama	Local models	✅ Built
Mistral	Small Latest	✅ Built
DeepSeek	Chat	✅ Built

Security Test Results

Integration Tests:    75/75   (100%) ✅
Advanced Security:    60/60   (100%) ✅
Red Team Tests:       16/16   (100%) ✅
OWASP LLM Top 10:    10/10   (100%) ✅
Total:               151 tests — all passing

What's Tested

Test Suite	Tests	Coverage
Capability Tokens	12	Wildcards, TTL, expiry
Credential Vault	8	Store, revoke, TTL
Skill Vetting	6	Static analysis, deps, sandbox
Input Sanitization	11	21 injection patterns
Filesystem Jail	12	Traversal, blocked paths
Network Egress	12	Allowlist, blocklist, ports
Risk Scoring	3	Normal → suspicious → compromised
AI Backend	4	Sanitize → AI → verify
Attack Simulation	7	Multi-stage attack chain
RAG Security	7	Poisoning, injection, trust
Privacy Filter	16	API keys, tokens, PII
Extraction Detection	9	Probing, system prompt theft
Hallucination	8	Overconfidence, impossible claims
Watermark	8	Stamp, verify, tamper detection
Kali Authorization	6	Target allowlist
Combined Attack	6	RAG + extraction + leak + tamper

12 HTTP Endpoints

# Start server
cargo run -- serve          # HTTP :8401
cargo run -- serve-tls      # HTTPS :8443

Endpoint	Method	Auth	Description
/health	GET	No	Health check
/api/v1/agents/run	POST	Admin, Operator	Run agent
/api/v1/agents/stop	POST	Admin, Operator	Stop agent
/api/v1/agents/list	GET	All	List agents
/api/v1/agents/status	GET	All	Agent status
/api/v1/inference	POST	Admin, Operator, Agent	AI inference
/api/v1/redteam	POST	Admin	Red team test
/api/v1/tools	POST	Admin, Operator, Agent	Execute tool
/api/v1/audit	GET	Admin, Operator, Viewer	Audit logs
/api/v1/metrics	GET	All	Prometheus metrics
/api/v1/dashboard	GET	All	Dashboard JSON
/mcp/tools/*	POST	MCP Auth	MCP tool calls

Quick Start

git clone https://github.com/Moudaxx/AEGIS-OS.git
cd AEGIS-OS
cp .env.example .env  # Add your API keys
cargo build
cargo run -- autonomous --cycles 3  # Watch it work alone!

Docker

docker-compose up -d  # AEGIS + Prometheus + Grafana

RBAC — 4 Roles

Role	Run	Stop	Inference	Red Team	Audit	Config
Admin	✅	✅	✅	✅	✅	✅
Operator	✅	✅	✅	❌	✅	❌
Viewer	❌	❌	❌	❌	✅	❌
Agent	❌	❌	✅	❌	❌	❌

AI Security Framework

Standard	Coverage
OWASP LLM Top 10	10/10
NIST AI RMF	Aligned
OWASP MAESTRO	Aligned

See AI-SECURITY-FRAMEWORK.md for full mapping.

Related Projects

• AEGIS Pay — Secure Payment Layer for AI Agents (x402 + EIP-3009 + USDC)

Tech Stack

Rust, Axum, Tokio, Wasmtime, Docker, Prometheus, Grafana, GitHub Actions

License

MIT — See LICENSE

Links

• 🌐 Website: moudaxx.github.io/AEGIS-OS
• 📄 Security Framework: AI-SECURITY-FRAMEWORK.md
• 💰 AEGIS Pay: github.com/Moudaxx/AEGIS-Pay