KOSMOS V2.0 Architecture
KOSMOS is an AI-Native Enterprise Operating System built on a multi-agent architecture with 11 specialized agents, 88 MCP tool servers, and enterprise-grade governance.
Design Principles
| Principle | Description |
|---|---|
| Agent-First | Every capability is an agent with tools, not monolithic code |
| Cost-Aware | Built-in governance for LLM spend ($50 auto, $50-100 vote, $100+ human) |
| Security-First | AEGIS has veto power over all actions |
| Observable | Full tracing from user request to tool execution |
| Cloud-Native | Kubernetes-native, GitOps deployment |
High-Level Architecture
Request Flow
Key Components
| Component | Description | Docs |
|---|---|---|
| Zeus | Master orchestrator - routes all requests, coordinates agents | Zeus |
| AEGIS | Security guardian - validates all actions, veto power | AEGIS |
| Pentarchy | 3-agent voting (Athena, Hephaestus, PROMETHEUS) for $50-$100 decisions | Governance |
| 11 Agents | Specialized domain agents (communication, scheduling, analytics, etc.) | Agents |
| 88 MCP Servers | Tool integrations across 9 domains | MCP Servers |
| PostgreSQL | Primary data store with pgvector, RLS, multi-tenancy | Schema |
Technology Stack
Documentation Sections
- System Topology - Detailed architecture with all components
- Governance - Pentarchy voting system and cost controls
- Security - 6-layer defense architecture and AEGIS
- Database Schema - Database schema and storage design
Auto-Generated Docs
Implementation details are auto-generated from source code:
- Agents - Python docstrings and methods
- API - OpenAPI schema
- MCP Servers - TypeScript tool definitions
- Configuration - Pydantic settings
- Database - SQL migrations