2LY Logo

Architecture

2LY under the hood

2LY connects AI agents to tools via MCP using NATS messaging, DGraph storage, and distributed runtimes.

Quick Architecture

Core Components

Agents

AI systems that connect via MCP protocol:

  • Claude Desktop
  • LangChain/LangGraph applications
  • Custom Python/TypeScript agents
  • n8n workflows

Role: Request and invoke tools for task execution

Runtime

Distributed execution environment (@2ly/runtime):

  • Agent Mode: Acts as MCP server for agents
  • Tool Mode: Spawns and manages MCP tool servers
  • Deployment: Local, cloud, or edge

Role: Bridge between agents and tools via NATS

Backend

Central orchestration service:

  • GraphQL API: Manage workspaces, runtimes, tools
  • Runtime Coordination: Track runtime health and route messages
  • Tool Discovery: Aggregate tool capabilities

Role: Coordinate distributed components and persist configuration

NATS

Message bus for all communication:

  • Pub/Sub: Real-time updates
  • Request/Reply: Tool execution
  • KV Store: Heartbeat and config

Role: Enable decoupled, scalable communication

DGraph

Graph database for platform data:

  • Workspaces and users
  • Runtimes and their configurations
  • MCP servers and discovered tools
  • Tool capabilities and assignments

Role: Persist and query platform state

Dashboard

React-based management interface:

  • Monitor runtime health
  • Configure MCP servers
  • Assign tool capabilities
  • View tool execution

Role: User interface for platform management

How It Works

  1. Agent connects to runtime via MCP
  2. Runtime registers with backend via NATS
  3. Tools are discovered and sent to agent
  4. Agent calls tool → NATS routes to tool runtime → executes → returns result

Why This Architecture?

Flexible Deployment

Run anywhere. Runtimes connect outbound to NATS, so they work:

  • Behind corporate firewalls (no inbound ports needed)
  • On edge devices with intermittent connectivity
  • In multiple cloud regions simultaneously
  • On developer laptops for testing

Geographic distribution becomes trivial—just deploy runtimes closer to your data or users.

High Availability

Message persistence means no lost requests during deployments or failures. JetStream stores messages until they're processed.

Automatic failover: If a runtime goes down, NATS routes to healthy instances automatically. No manual intervention, no downtime.

Zero-downtime deployments: Update services without interrupting active tool calls. The message bus handles routing during transitions.

Observability Built-In

See everything. Track every tool call, monitor runtime health, debug failures—all from the dashboard.

Enterprise-ready: Audit trails, usage analytics, and debugging tools included. No need for third-party monitoring services for basic observability.

For more details, see the Why 2LY section in Installation.

Deployment

Run locally

Start all 2LY services on localhost via Docker Compose:

docker compose up -d

Access at http://localhost:8888. See Installation for details.

Production

You can deploy 2LY using official container images:

See docker-compose.yml for an example.

Next Steps

Deploy and Run

Deploy your configured toolset and link them to agents through the 2LY web interface

Agents

Understanding agents in 2LY