LIATE — L[LM] · I[ntegration] · A[gent] · T[ools] · E[nvironment]

Class is an Agent. Agent is an API.

The Open Source AI Agent Runtime for OOP Architecture. 🇮🇳

🎉 Liate is selected for the Sarvam AI Startup Program.
Built solo by Vinod Hatti (@solovinod)

Run Millions of AI Agents on Edge Cloud. Free Forever. Self Host.
Agent is a Function on Neon — 1,000,000 free runs/month on your own account.

If your language has a class — your class can be an AI Agent.
Java · JavaScript · TypeScript · Python

Run Locally

Deploy Free — 1M Agent Runs/Month (BYOC)

---

Not a Framework. A Runtime.

Frameworks tell you how to write code.
A runtime executes your code.

Liate is the runtime for AI Agents.
You write a class. Liate runs it as an autonomous AI Agent — locally, on edge, or on your own cloud.

---

The 3 Pillars

• 🛠️ Build Your Own AI Agent — Extend LiateAgent. Your class IS the agent and the API. Zero boilerplate.
• 🧠 Bring Your Own AI (BYOAI) — Plug in your own Sarvam AI key (SARVAM_API_KEY). Direct access, zero markup.
• ☁️ Bring Your Own Cloud (BYOC) — Connect your own Neon account. 1,000,000 free agent runs/month on your own infrastructure.

---

Hello Agent ⚡

The frontier agent. One class. All 5 layers. Any OOP language.

TypeScript

import { LiateAgent } from "liate";

class HelloAgent extends LiateAgent {
  constructor() {
    super({
      L: "sarvam/sarvam-105b",
      I: { memory: "sessions/hello" },
      A: { name: "hello-agent", prompt: "You are HelloAgent." },
      T: ["mcp/local"],
      E: { SARVAM_API_KEY: process.env.SARVAM_API_KEY }
    });
  }
}

await new HelloAgent().run("Hello Agent World!");

Java

import dev.liate.LiateAgent;

public class HelloAgent extends LiateAgent {
    public HelloAgent() {
        super(new AgentConfig()
            .L("sarvam/sarvam-105b")
            .I("sessions/hello")
            .A("hello-agent", "You are HelloAgent.")
            .T("mcp/local")
            .E("SARVAM_API_KEY")
        );
    }
}

new HelloAgent().run("Hello Agent World!");

Python

from liate import LiateAgent
import os

class HelloAgent(LiateAgent):
    def __init__(self):
        super().__init__(
            L="sarvam/sarvam-105b",
            I={"memory": "sessions/hello"},
            A={"name": "hello-agent", "prompt": "You are HelloAgent."},
            T=["mcp/local"],
            E={"SARVAM_API_KEY": os.environ["SARVAM_API_KEY"]}
        )

await HelloAgent().run("Hello Agent World!")

JavaScript

import { LiateAgent } from "liate";

class HelloAgent extends LiateAgent {
  constructor() {
    super({
      L: "sarvam/sarvam-105b",
      I: { memory: "sessions/hello" },
      A: { name: "hello-agent", prompt: "You are HelloAgent." },
      T: ["mcp/local"],
      E: { SARVAM_API_KEY: process.env.SARVAM_API_KEY }
    });
  }
}

await new HelloAgent().run("Hello Agent World!");

Or zero-code with agent.json:

{
  "L": "sarvam/sarvam-105b",
  "I": { "memory": "sessions/hello" },
  "A": { "name": "hello-agent", "prompt": "You are HelloAgent." },
  "T": ["mcp/local"],
  "E": { "SARVAM_API_KEY": "$SARVAM_API_KEY" }
}

npx liate run agent.json "Hello Agent World!"

---

Run on the fly

No project setup needed:

npx liate run "Analyze this repository and find performance bottlenecks"

---

Deploy to Your Own Cloud (BYOC) 🚀

Deploy the Liate runtime engine directly into your own Neon account:

# Connect your own Neon account via OAuth
liate connect --neon

# Deploy Liate engine to your Neon account
liate deploy --neon

• Scale to Zero — zero cost when agents sleep.
• Neon Cron — scheduled agents wake up automatically.
• 1,000,000 free executions/month on your own account.
• Your data never leaves your own infrastructure.

---

The LIATE Architecture

Every agent runs on 5 layers, declared as class properties:

+--------------------------------------------------+
|  [L] LLM           — Sarvam AI (India-First)     |
+--------------------------------------------------+
|  [I] Integration   — Conversation Memory         |
+--------------------------------------------------+
|  [A] Agent         — Name + Intent               |
+--------------------------------------------------+
|  [T] Tools         — MCP Server Connector        |
+--------------------------------------------------+
|  [E] Environment   — API Keys & Edge Runtime     |
+--------------------------------------------------+

The 5 Layers (L·I·A·T·E)

Layer	Name	Day 1	Community PRs
L	LLM	Sarvam AI (hub/llm/sarvam.json) 🇮🇳	Add OpenAI, Claude, Gemini...
I	Integration	In-memory conversation state	Add Neon Postgres, Redis...
A	Agent	LiateAgent class — name, prompt, intent	Add ReAct, multi-agent...
T	Tools	Local MCP v2 (stdio)	Add remote MCP servers...
E	Environment	Pure .env key reader	Add edge isolate configs...

---

Contributing

Liate's core runtime is minimal and focused.
The hub/ directory belongs to the community:

• Add a new LLM → PR to hub/llm/
• Share your agent → PR to hub/agents/

---

License

MIT © 2026 Vinod Hatti (@solovinod). Free and open source forever.