The Doctrine - OpenLiate

The Four Truths

OpenLiate is structured around four fundamental architectural principles:

Class is an Agent.
Agent is a Function.
Function is an API.
LAPI is Your Backend.

1. Class is an Agent

In classical OOP, a class encapsulates state, properties, and methods. In OpenLiate, an Agent is an instantiated class where:

Properties: Configuration defined by the L · I · A · T · E layers in agent.json.
State: Session history, scratchpad notes, observation memory, and step counters.
Methods: Execution loops (run), tool dispatchers (executeTool), and memory persistence.

// 1 Class = 5 Properties (L · I · A · T · E) = Same as agent.json
class HelloAgent {
  L; // Model
  I; // Knowledge & Memory
  A; // Harness & ReAct Loop
  T; // Tools & MCP
  E; // Environment & Sandbox
}

2. Agent is a Function

When invoked, the entire agent class collapses into a single, callable, pure function. It accepts an input prompt and optional session ID, executes the autonomous loop, and returns a structured output:

import { Agent } from "openliate";

const agent = new Agent({
  manifest: "agent.json"
});

const result = await agent.run({
  prompt: "Verify taxpayer PAN AABCC1234D",
  sessionId: "0192e3a1-7b82-7f91-8abc-123456789abc"
});

console.log(result.output);

3. Function is an API

Every agent function can be seamlessly exported as a network service without writing boilerplate router code. By passing the agent to the LAPI server, the function is automatically mounted as an HTTP microservice with built-in CORS, error handling, and parameter validation.

4. LAPI is Your Backend

The Liate Agent Protocol Interface (LAPI) acts as the headless backend for modern AI applications. Frontends (Next.js, React, mobile apps, or voice clients) interact directly with LAPI endpoints:

GET /health — Verify engine readiness and active doctrine.
GET /agents — Query the agent registry in the local Hub.
POST /run — Execute any agent dynamically with UUIDv7 monotonic session tracking.