Agents

Agents are AI systems composed of an LLM, prompts, and tools. 2LY decouples tools from agent logic through a gateway runtime, enabling centralized governance and scalable tool management across all agents.

What are Agents?

Traditional agents bundle three components:

• LLM: Language model for reasoning (GPT-4, Claude, Llama)
• Prompts: Instructions defining agent behavior
• Tools: Functions for external actions (APIs, databases, files)

2LY separates the tool layer from agent frameworks. Instead of integrating tools directly into each agent, agents connect as MCP clients to a runtime that manages tool access, authentication, and execution centrally. This architecture allows you to configure tools once and use them across any agent framework.

How Agents Work

1. Runtime starts and registers with backend via NATS
2. Backend sends available tools to runtime
3. Agent lists tools from runtime
4. Agent calls tool → routed via NATS → executes → returns result

Benefits

• Dynamic Discovery: Add/remove tools without changing agent code
• Centralized Config: Manage tools in dashboard, not in code
• Distributed: Access tools anywhere (local, cloud, edge)
• Framework Agnostic: Works with any MCP-compatible framework

Next Steps

• Connect Agent - Step-by-step guide
• Tools - What tools can do
• Runtime - How tools execute