Kubernetes for AI Agents. Discover, coordinate, and orchestrate any AI agent from any vendor — seamlessly.
AgentOS is a vendor-neutral, protocol-first platform that lets AI agents — regardless of framework (LangChain, CrewAI, OpenAI, custom) — discover each other, communicate, and execute multi-step workflows together. It provides the missing coordination layer for multi-agent systems.
| Problem | AgentOS Solution |
|---|---|
| Agents from different vendors can't talk to each other | Protocol bridges (MCP, A2A, OpenAI) make any agent a first-class citizen |
| No standard way to discover agent capabilities | Agent Registry with capability matching and trust scoring |
| Multi-agent workflows are fragile | DAG-based workflows with circuit breakers, failover, and dead-letter queues |
| No visibility into what agents are doing | Real-time dashboard, audit trails, cost tracking, Prometheus metrics |
| Building multi-agent systems requires glue code | TypeScript + Python SDKs, CLI, 6 workflow templates, ready in minutes |
# 1. Clone and install
git clone https://github.com/IntellectAI/agentos.git
cd agentos && npm install && npm run build
# 2. Start infrastructure (NATS, Redis, PostgreSQL)
docker compose -f infra/docker-compose.yml up -d
# 3. Start all services (Registry, Orchestrator, Dashboard, Platform)
npm run dev:registry # :3100
npm run dev:orchestrator # :3200
# Open http://localhost:3300 for dashboard
# 4. Register an agent
npx aos agent register --name "my-summarizer" --actions summarize
# 5. Run a workflow
npx aos run --template research-and-report --set-QUERY "AI agent trends"Or install the CLI globally:
npm install -g @agentos/cli
aos init my-project && cd my-project && aos deploy┌──────────────────────────────────────────────────────────────────┐
│ Layer 4: Governance & Observability │
│ Dashboard · Approval Gates · Audit Trail · Cost Tracker │
│ Prometheus Metrics · Correlation IDs · Structured Logging │
├──────────────────────────────────────────────────────────────────┤
│ Layer 3: Communication & State Bus │
│ NATS Message Bus · Redis State Store · Context Manager │
│ Circuit Breaker · Agent Pool · Failover · Dead Letter Queue │
├──────────────────────────────────────────────────────────────────┤
│ Layer 2: Task Decomposition & Planning │
│ Goal Parser · DAG Builder · Agent Matcher · LLM Adapters │
│ Capability Validator · Workflow Templates │
├──────────────────────────────────────────────────────────────────┤
│ Layer 1: Agent Registry & Discovery │
│ Registry Service · Capability Schema · Health Monitor │
│ Trust Scoring · Protocol Bridges (MCP, A2A, OpenAI) │
└──────────────────────────────────────────────────────────────────┘
TypeScript:
import { AgentOSClient } from '@agentos/sdk';
const client = new AgentOSClient({ registryUrl: 'http://localhost:3100/api/v1' });
await client.registerAgent({ name: 'my-agent', actions: ['analyze'] });
const plan = await client.planGoal('Research AI trends and summarize findings');Python:
from agentos import AgentOSClient, BaseAgent
async with AgentOSClient(registry_url="http://localhost:3100/api/v1") as client:
await client.register_agent(name="my-agent", actions=["analyze"])Platform API:
import { PlatformClient } from '@agentos/sdk';
const platform = new PlatformClient({ baseUrl: 'http://localhost:3400' });
const { token } = await platform.login({ email: 'you@company.com', password: '...' });
const listings = await platform.searchListings({ category: 'research' });Core Orchestration — Agent registration & discovery via AOP protocol, DAG-based workflow planning with LLM-powered goal parsing, NATS messaging + Redis state, circuit breaker, agent pooling, failover, dead-letter queues.
LLM Integration — Anthropic Claude + OpenAI GPT providers, per-token cost tracking with budget enforcement, graceful fallback to pattern-based logic.
Protocol Bridges — MCP (Anthropic Model Context Protocol), A2A (Google Agent-to-Agent), OpenAI Function Calling.
Platform — Multi-tenant architecture, RBAC + JWT auth, SSO (SAML 2.0 + OIDC), Stripe billing, agent marketplace, visual workflow builder.
Infrastructure — Docker Compose (7 services) + Kubernetes manifests, CI/CD (GitHub Actions), Prometheus metrics, correlation IDs, structured logging, graceful shutdown.
Developer Experience — TypeScript + Python SDKs, CLI (12 commands), 6 workflow templates, 3 end-to-end demos, comprehensive docs.
agentos/
├── packages/ # 15 packages: core, registry, planner, state, resilience,
│ # sdk-ts, sdk-py, cli, auth, billing, marketplace,
│ # workflow-builder, hosted-registry, database, integration
├── agents/ # 3 LLM-powered agents: researcher, coder, summarizer
├── bridges/ # 3 protocol bridges: MCP, A2A, OpenAI
├── dashboard/ # Monitoring dashboard (API + UI)
├── templates/ # 6 workflow templates
├── examples/ # 3 end-to-end demos (TS, Python, multi-agent)
├── docs/ # AOP spec, architecture, getting started
├── infra/ # Docker Compose, Dockerfiles (4), K8s manifests (11)
└── website/ # Documentation site
| Component | Technology |
|---|---|
| Core Runtime | TypeScript / Node.js 20 |
| Agent Adapters | Python SDK (async, Pydantic) |
| Message Bus | NATS (JetStream) |
| State Store | Redis |
| Metadata DB | PostgreSQL (8 migrations) |
| API | REST + WebSocket |
| LLM | Anthropic Claude + OpenAI GPT |
| Auth | JWT (HS256) + SAML/OIDC SSO |
| Billing | Stripe |
| Infrastructure | Docker Compose → Kubernetes |
| CI/CD | GitHub Actions |
| Observability | Prometheus + Pino + Correlation IDs |
| Free | Pro | Enterprise | |
|---|---|---|---|
| Price | $0 | $49/mo | Custom |
| Agents | 5 | 25 | Unlimited |
| Operations/mo | 1,000 | 10,000 | Unlimited |
| Team Members | 3 | 20 | Unlimited |
| SSO | — | — | SAML + OIDC |
| Support | Community | Priority | Dedicated + SLA |
- Getting Started Guide — Setup to first workflow in 30 minutes
- AOP Specification v0.1 — Full protocol reference (12 sections)
- Architecture Overview — 4-layer system design
- Multi-Tenant Architecture — Org isolation, RBAC, SSO
- Architecture Diagrams — 6 Mermaid diagrams
- Decision Records — ADR-001 through ADR-007
- Changelog — Release history
- Contributing — How to contribute
npm install # Install all dependencies
npm run build # Build all 22 packages
npm test # Run 1,075 tests
npm run lint # Lint all packages
npm run docker:up # Start Docker infrastructure
npm run docker:down # Stop Docker infrastructureBuilt by IntellectAI — Making multi-agent AI accessible to everyone.