Autonomous Agent Operating System (AOS)

Enterprise Control Plane for AI Agents — Governance · Orchestration · Observability · Billing

AOS is the infrastructure layer that sits above AI agent frameworks. It does not replace LangGraph, CrewAI, or Agent Swarm — it governs them. Every agent dispatch is policy-checked, metered, traced, and auditable.

---

Documentation

Full documentation →

Guide	Description
Architecture	How the control plane and execution engine fit together
Setup & Installation	Prerequisites, environment config, first-time startup
Usage Guide	Running agents, using the API, real examples
API Reference	All 50+ REST endpoints with request/response schemas
Agent Catalogue	All 245 swarm agents organized by category
Policy Engine	Writing governance rules, conditions, compliance templates
Swarm Bridge	How AOS and Agent Swarm communicate
Billing & Metering	Cost tracking, budgets, department chargeback
Observability	Traces, audit logs, Prometheus metrics
Security	Auth, RBAC, encryption, production cautions
Configuration	All environment variables and settings
Contributing	Adding agents, skills, API modules

---

What Is AOS?

┌─────────────────────────────────────────────────────────────────┐
│                  AOS Django Backend (Port 8000)                 │
│                                                                 │
│  Agent IAM & Registry    Policy & Governance Engine             │
│  Usage Metering          Observability & Audit Logs             │
│  Knowledge Base (RAG)    Swarm Bridge (/api/swarm/)             │
└───────────────────────────────┬─────────────────────────────────┘
                                │ HTTP / JWT
┌───────────────────────────────▼─────────────────────────────────┐
│              Agent Swarm (245+ Specialized Agents)              │
│                                                                 │
│  Engineering · Marketing · Sales · Finance · Compliance         │
│  Product · Strategy · Design · QA · Support · Game Dev          │
│                                                                 │
│  Self-Healing · Engine-Agnostic (Claude / Gemini / Custom)      │
│  5-Phase Workflow · Cross-Session Memory                        │
└─────────────────────────────────────────────────────────────────┘

AOS provides:

• Agent Identity & IAM — unique cryptographic identity, JWT sessions, RBAC roles
• Policy Enforcement — declarative rules with 10 operators, 4 effects (ALLOW/DENY/AUDIT/ESCALATE)
• Orchestration — multi-agent supervisor graphs via LangGraph
• Observability — full execution traces, immutable audit logs, Prometheus metrics
• Billing — token + compute cost tracking with department-level chargeback

Agent Swarm provides:

• 245 specialized agents across 20 business domains
• 5-phase autonomous workflow (questionnaire → plan → execute → debug → ship)
• Self-healing with 5 recovery strategies (retry, reassign, simplify, fallback, escalate)
• LLM engine agnosticism (Claude, Gemini, or custom CLI)
• Cross-session memory with compressed learnings

---

Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+
• Git, curl

One-command start

git clone <repo-url> Agentic-Enterprise
cd Agentic-Enterprise

# Add your API keys to backend/.env
echo "ANTHROPIC_API_KEY=sk-ant-..." >> backend/.env
echo "GEMINI_API_KEY=AIza..."     >> backend/.env

# Start everything
./start.sh

start.sh handles migrations, policy seeding, agent sync, JWT provisioning, and a smoke test automatically. See Setup Guide for full details.

Run a governed agent

cd agent-swarm
source .env
python orchestrator.py "Research top 10 enterprise fintech prospects"

Every dispatch is automatically policy-checked, token-metered, and trace-logged by AOS.

---

Services

Service	URL	Credentials
REST API	http://localhost:8000	JWT via /api/token/
Swagger UI	http://localhost:8000/api/docs/swagger/	—
Admin Panel	http://localhost:8000/admin/	admin / admin1234
Prometheus	http://localhost:8000/metrics	—

---

System at a Glance

Metric	Value
Swarm agents	245
Agent categories	20
REST API endpoints	50+
Policy condition operators	10
LLM providers supported	6
Self-healing strategies	5

---

Manage the System

./start.sh              # Start everything
./start.sh --stop       # Kill all managed processes
./start.sh --status     # Show running PIDs

# Backend management (from backend/ with .venv active)
python manage.py sync_swarm_agents     # Sync 245 agents into AOS registry
python manage.py default_policies      # Seed default governance policies
python manage.py migrate               # Apply database migrations
python manage.py createsuperuser       # Create admin user

---

Architecture Summary

The system consists of two components that operate together:

backend/ — Django REST API serving as the control plane:

• apps/agent_registry/ — Agent identity and metadata
• apps/agent_gateway/ — Authentication and request auditing
• apps/policy_engine/ — Declarative governance and rule evaluation
• apps/agent_intelligence/ — LangGraph execution and trace logging
• apps/knowledge_base/ — ChromaDB RAG and document management
• apps/billing/ — Usage metering and cost attribution
• apps/swarm_bridge/ — Integration bridge to Agent Swarm

agent-swarm/ — Multi-agent execution engine:

• orchestrator.py — 5-phase workflow dispatcher
• core/aos_client.py — HTTP client connecting swarm to AOS bridge
• core/self_healer.py — Automatic failure recovery
• agents/ — 245 specialized agent definitions
• skills/ — 239 reusable knowledge modules
• commands/ — 125 workflow templates

Full architecture: docs/architecture.md

---

Security Notice

Before deploying outside localhost:

• Change SECRET_KEY in backend/backend/settings.py
• Set DEBUG=False
• Change the admin password (admin1234)
• Use PostgreSQL instead of SQLite
• Enable HTTPS via a reverse proxy

Full security guide: docs/security.md

---

License

See LICENSE