Metadata-Version: 2.5
Name: openliate
Version: 0.1.1
Summary: Open Source Backend Layer for AI Agents — Built for Sovereign AI Agents From India to the World
Project-URL: Homepage, https://openliate.tech
Project-URL: Repository, https://github.com/openliate/openliate
Project-URL: Documentation, https://docs.openliate.tech
Project-URL: Issues, https://github.com/openliate/openliate/issues
Author-email: OpenLiate <vinod@openliate.tech>, Vinod Hatti <gitdevinod@gmail.com>
License: MIT
Keywords: agent,aum,autonomous-agents,liate,mcp,om,sarvamai,sovereign-ai
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Classifier: Programming Language :: Python :: 3.15
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Requires-Python: >=3.9
Requires-Dist: httpx>=0.24.0
Description-Content-Type: text/markdown

<div align="center">

# OpenLiate — Open Source Backend Layer for AI Agents
### Built for Sovereign AI Agents From India to the World

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

<br>

<p align="center">
  <a href="https://docs.openliate.tech"><img src="https://img.shields.io/badge/docs-docs.openliate.tech-white?labelColor=black&style=flat-square&v=2" alt="Docs"></a>
  <a href="https://github.com/openliate/openliate/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-white?labelColor=black&style=flat-square" alt="License: MIT"></a>
  <a href="https://sarvam.ai"><img src="https://raw.githubusercontent.com/OpenLiate/OpenLiate/main/assets/sarvam-badge.svg" alt="Startup Program: Sarvam AI"></a>
  <a href="https://neon.tech"><img src="https://img.shields.io/badge/Neon-Cloud_%26_AI_Gateway_Credits-white?logo=neon&logoColor=00E59B&labelColor=black&style=flat-square" alt="Neon Cloud & AI Gateway Credits"></a>
  <a href="https://codespaces.new/openliate/openliate"><img src="https://img.shields.io/badge/Codespaces-1--Click_Run-white?logo=github&logoColor=white&labelColor=black&style=flat-square" alt="Open in GitHub Codespaces"></a>
  <img src="https://img.shields.io/badge/version-v0.1.0-white?labelColor=black&style=flat-square" alt="v0.1.0">
  <img src="https://img.shields.io/badge/Bun-Supported-white?logo=bun&logoColor=white&labelColor=black&style=flat-square" alt="Bun">
  <img src="https://img.shields.io/badge/Node.js-Supported-white?logo=node.js&logoColor=5FA04E&labelColor=black&style=flat-square" alt="Node.js">
  <img src="https://img.shields.io/badge/Deno-Supported-white?logo=deno&logoColor=white&labelColor=black&style=flat-square" alt="Deno">
</p>

<p align="center">
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Regional_Cloud-2.5M_Live-white?labelColor=black&style=flat-square" alt="Regional Cloud"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Neon-1M_Runs-white?logo=neon&logoColor=00E59B&labelColor=black&style=flat-square" alt="Neon"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Vercel-1M_Runs-white?logo=vercel&logoColor=white&labelColor=black&style=flat-square" alt="Vercel"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Supabase-5_Lakhs-white?logo=supabase&logoColor=3ECF8E&labelColor=black&style=flat-square" alt="Supabase"></a>
</p>

<p align="center">
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Worldwide_Edge-16M_Live-white?labelColor=black&style=flat-square" alt="Worldwide Edge"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Fastly-10M_Runs-white?logo=fastly&logoColor=FF282D&labelColor=black&style=flat-square" alt="Fastly"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Cloudflare-3M_Runs-white?logo=cloudflare&logoColor=F38020&labelColor=black&style=flat-square" alt="Cloudflare"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/Azion-3M_Runs-white?logo=data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHZpZXdCb3g9IjAgMCAyNCAyNCI+PHBhdGggZmlsbD0iI0Y0NjUyQiIgZD0iTSAyIDE5LjMgTCAxOS44IDIgTCAyMiAyIEwgMTguOCAyMiBMIDE1LjUgMjIgTCAxNy42IDguNyBMIDQuMiAyMiBMIDIuOCAyMiBMIDIgMjEuMSBaIi8+PC9zdmc+&labelColor=black&style=flat-square" alt="Azion"></a>
</p>

<p align="center">
  <a href="#aum-architecture"><img src="https://img.shields.io/badge/Agent_Architecture-AUM_(Agent_Unified_Manifest)-white?labelColor=black&style=flat-square" alt="Agent Unified Manifest (AUM)"></a>
  &nbsp;
  <a href="#runtime-engine"><img src="https://img.shields.io/badge/Core_Engine-OM_(Orchestration_Module)-white?labelColor=black&style=flat-square" alt="Core Engine: OM (Orchestration Module)"></a>
  &nbsp;
  <a href="#instant-install"><img src="https://img.shields.io/badge/Instant_Install-Zero_Dependencies-white?labelColor=black&style=flat-square" alt="Instant Install"></a>
  &nbsp;
  <a href="#run-cloud"><img src="https://img.shields.io/badge/AI_Gateway-Neon_AI_Gateway-white?logo=neon&logoColor=00E59B&labelColor=black&style=flat-square" alt="Neon AI Gateway"></a>
  &nbsp;
  <a href="https://skills.sh"><img src="https://img.shields.io/badge/Agent_Skills-1.26M%2B_skills.sh_(Live)-white?labelColor=black&logo=vercel&logoColor=white&style=flat-square" alt="1.26M+ skills.sh"></a>
</p>

<br>

> **#1 Milestone: Liate — Selected for India's Sarvam AI Startup Program 2026** <sup>IN</sup>  
> Open Source Backend Layer for AI Agents by [OpenLiate](https://github.com/openliate) · Creator [@gitdevinod](https://github.com/gitdevinod)
> 
> **#2 Milestone: Neon Cloud Developer Runway — Compute & AI Gateway Credits** <sup>CLOUD</sup>  
> Powering serverless Postgres compute, 63-model AI Gateway routing, and sovereign agent execution · [Neon Cloud](https://neon.tech)

<br>

> ### *"If Stripe did payments in 7 lines of code, why not AI Agents in 5 lines of code?*  
> **BaaA — Backend-as-an-Agent.**  
> *Let's Try Liate."*

</div>

## <a id="instant-install"></a>[0] Instant Install (Zero Dependencies)

> [!NOTE]
> **CLI Command Clarity (`openliate` & `liate`):**  
> Both the full canonical binary **`openliate`** and the fast punchy alias **`liate`** are installed together and work interchangeably across all commands:
> - `openliate run agent.json` = `liate run agent.json`
> - `openliate add skills <name>` = `liate add skills <name>`
>
> Feel free to use the short, punchy `liate` in your daily terminal workflow, or `openliate` in CI/CD pipelines and scripts.

```bash
# 1-Click Run via npx (Zero local install):
npx openliate run agent.json "Hello World"

# Or Install Globally:
npm install -g openliate

# Both commands are immediately available:
liate run agent.json "Hello World"
openliate run agent.json "Hello World"

# Windows (PowerShell):
powershell -c "irm https://openliate.tech/install.ps1 | iex"

# macOS / Linux (cURL):
curl -fsSL https://openliate.tech/install.sh | sh
```

> **Prefer 1-Click in Browser?**  
> Run OpenLiate instantly inside [GitHub Codespaces](https://codespaces.new/openliate/openliate) with zero local setup.


> **Why 5 Lines? The AI Agent Architecture:**  
> An autonomous AI Agent is fundamentally 5 orthogonal layers: **LLM (`L`)**, **Integration (`I`)**, **Agent (`A`)**, **Tools (`T`)**, and **Environment (`E`)**. OpenLiate maps this directly into 5 lines of declarative execution.

```jsonc
{
  "L": "sarvam/sarvam-105b",                                                          // [L] LLM: Sarvam AI Token Factory (2026) & Neon AI Gateway (63+ Models)
  "I": { "memory": "sessions/hello-agent" },                                          // [I] Integration: Session Memory, State, npm Libraries, Git Repos
  "A": { "name": "hello_agent", "intent": "Autonomous Agent", "skills": ["vercel/nextjs"] }, // [A] Agent: Identity, Intent, Vercel skills.sh, ReAct Loop
  "T": ["fetch"],                                                                     // [T] Tools: Model Context Protocol (MCP v2 Local Stdio & Remote SSE)
  "E": { "SARVAM_API_KEY": "$SARVAM_API_KEY" }                                        // [E] Environment: Secrets & Security Config
}
```

```bash
# Execute your agent (reads L·I·A·T·E directly from agent.json)
liate run agent.json "Hello Agent World!"
```

> **Prefer writing code directly?**  
> Direct language SDKs for **TypeScript / JavaScript** (`npm install openliate`) and **Python** (`pip install openliate`) are live on **v0.1.0**.  
> Write your agent in 5 properties: **`class extends liate`**.

<br>

---

<div align="center">

### Millions of AI Agent Runs on Edge Cloud: Function is an Agent.
#### *Self-Host Anywhere or Bring Your Own Cloud (BYOC) — Free Forever.*

<br>

## **Class is an Agent. Agent is a Function. Function is an API.**

#### *Liate Agent Protocol Interface (LAPI) is Your Backend.*

</div>

<br>

Write your sovereign agent directly in code. OpenLiate maps the 5 orthogonal layers (`L · I · A · T · E`) into clean OOP classes—available interchangeably in shorthand (`L, I, A, T, E`) or full-name (`LLM, Integration, Agent, Tools, Environment`):

<br>

#### 1. TypeScript & JavaScript SDK (`npm install openliate`)

```typescript
// agent.ts
import { liate } from "openliate";

export class HelloAgent extends liate {
  L = "sarvam/sarvam-105b";                                                          // [L] LLM: Sarvam AI Token Factory (2026) & Neon AI Gateway (63+ Models)
  I = { memory: "sessions/hello-agent" };                                             // [I] Integration: Co-located Session State
  A = { name: "hello_agent", intent: "Autonomous Sovereign Agent" };                 // [A] Agent: Identity & ReAct Loop
  T = ["fetch"];                                                                     // [T] Tools: Model Context Protocol (MCP v2)
  E = { SARVAM_API_KEY: "$SARVAM_API_KEY" };                                         // [E] Environment: Secrets & API Keys
}

// Run in-process:
const agent = new HelloAgent();
const result = await agent.run("Hello Agent World!");
console.log(result.text);

// Or convert to a standard Web API handler (Hono / Vercel / Cloudflare):
export const POST = agent.toHandler();
```

```bash
# Execute your TypeScript agent directly:
bun run agent.ts
```

<br>

#### 2. Python SDK (`pip install openliate`)

```python
# agent.py
from openliate import liate

class HelloAgent(liate):
    L = "sarvam/sarvam-105b"                                                          # [L] LLM: Sarvam AI Token Factory (2026) & Neon AI Gateway (63+ Models)
    I = {"memory": "sessions/hello-agent"}                                            # [I] Integration: Co-located Session State
    A = {"name": "hello_agent", "intent": "Autonomous Sovereign Agent"}               # [A] Agent: Identity & ReAct Loop
    T = ["fetch"]                                                                     # [T] Tools: Model Context Protocol (MCP v2)
    E = {"SARVAM_API_KEY": "$SARVAM_API_KEY"}                                        # [E] Environment: Secrets & API Keys

# Run via LAPI engine:
agent = HelloAgent()
result = agent.run("Hello Agent World!")
print(result.response)
```

```bash
# Execute your Python agent directly:
python agent.py
```

> **Full-Name Support:** Prefer self-documenting property names? Both SDKs and `agent.json` natively accept:  
> `LLM` · `Integration` · `Agent` · `Tools` · `Environment` (or lowercase `llm`, `integration`, `agent`, `tools`, `environment`).

---

## <a id="runtime-engine"></a>[1] Core Engine: OM (Orchestration Module)

**Liate's Orchestration Module (OM) is the autonomous runtime engine** — just like Docker executes containers and Ollama executes models, **Liate executes AI agents via OM** across the 5 orthogonal layers (`L · I · A · T · E`).

Liate provides three sovereign deployment targets across local hardware, regional cloud, and worldwide edge:

---

### #1 Run Locally (Native CLI & 1-Click Docker Engine)

Run agents anywhere on your own hardware with full Dual-MCP superpowers:

```bash
# 1. Native Standalone CLI
liate run agent.json "Hello Agent World!"

# 2. Local AI Gateway & Engine Server (OpenAI-Compatible Local LAPI Gateway /v1)
openliate serve --port 3000

# Connect Cursor, Claude Code, Cline, Copilot, or Python SDK to your local port:
export OPENAI_BASE_URL="http://localhost:3000/v1"
export OPENAI_API_KEY="any-key"

# 3. Production Docker Engine (Node 24 + Python 3.14 + Astral uv + npx)
docker compose up -d
```

* **Full Dual-MCP Support:** Execute both local Stdio MCP tools (`npx`, `uvx`) and remote HTTP/SSE MCP endpoints.
* **100% Offline & Private:** Zero cloud dependencies, zero external network hops, zero telemetry.
* **Your Keys:** Reads `$SARVAM_API_KEY` (or OpenAI/Claude) directly from your local `.env`.
* **Universal Deployment:** Run on your laptop, Mac/Linux/Windows, any VPS (DigitalOcean, Hetzner, AWS), or private Kubernetes cluster.

---

### #2 Run in Regional Cloud (Co-Located Postgres Memory · 2.5M Free Runs/mo)

Link your Vercel, Neon, and Supabase accounts with Liate in one click to deploy your autonomous agent engine across all three for `< 1ms` database retrieval:

```bash
# 1-Click Link & Auto-Deploy across all 3 platforms (2.5M Free Runs/mo for $0.00)
liate link

# Or link individual clouds:
liate link --neon             # Provisions project 'liate' & deploys engine (2 vCPU, 8GB RAM + pgvector)
liate link --vercel           # Deploys serverless HTTP engine (1M runs, auto-scales to 0)
liate link --vercel-daemons   # Deploys 24/7 containerized background daemon worker (Node 24 + Python + uvx)
liate link --supabase         # Deploys Edge Function (500k runs + pgvector store)

# Relink / refresh tokens and redeploy anytime:
liate relink                  # Relink all clouds with fresh OAuth
liate relink --supabase       # Relink specific provider

# Unlink & clear saved credentials:
liate unlink --all            # Safely clear all tokens from disk and .env.local

# Optional: set private model keys
liate key set SARVAM_API_KEY="your-key-here"
liate key list
```

* **Neon Functions (1,000,000 / mo):** Linux Node.js (**2 vCPU, 8 GB RAM**) co-located directly with serverless Postgres + `pgvector` and automated Neon Cron triggers. Plus **all 13 Context.dev serverless modules** running natively on Neon compute.
* **Vercel Functions & Daemons (1,000,000 / mo):** Fluid Compute functions for Next.js apps, plus 24/7 background agent runners via October 15, 2026 **Vercel Daemons** (`Dockerfile`).
* **Supabase Edge (500,000 / mo):** Deno V8 edge functions co-located with Postgres auth and realtime database webhooks.
* **Zero Cross-Region Latency:** Compute lives in the same data center as your database—eliminating network bottlenecks.

---

### #3 Run in Worldwide Edge Cloud (Anycast BGP Edge Fleet · Live)

Distribute agents across 300+ to 500+ global Points of Presence (PoPs) using Anycast BGP routing for sub-10ms response times worldwide:

```bash
# Link your edge cloud providers:
liate link --cloudflare
liate link --azion
liate link --fastly
```

* **Cloudflare Workers (3,000,000 / mo · Live):** Global Google V8 Isolates across 300+ edge cities (100k requests/day free).
* **Azion Edge Functions (3,000,000 / mo · Live):** Distributed V8 Isolates across 500+ global edge locations with edge storage caching and OAuth MCP integration.
* **Fastly Compute (10,000,000 / mo · Live):** Pure WebAssembly (WASM) on Wasmtime/Lucet with microsecond cold starts. Orchestrated via Fastly's `@fastly/mcp` streamable-http remote adapter.
* **Global Low Latency:** Requests terminate at the physical data center closest to the user or webhook sender.

---

### <a id="run-cloud"></a>Multi-Cloud Capacity Matrix (18.5M Live Runs/mo · Zero Vendor Lock-in)

OpenLiate organizes multi-cloud agent execution across two clean physical topologies: **Regional Cloud** (co-located with your Postgres memory) and **Worldwide Edge Cloud** (sub-10ms global ingress):

| Layer | Physical Topology | Platforms | Free Monthly Runs | Compute Specs & Architecture | Tool Transport `[T]` | Native Memory `[I]` | Cost |
| :---: | :--- | :--- | :---: | :--- | :--- | :--- | :---: |
| **Local** | **Self-Host Hardware** | **Local / VPS Docker** *(Live)* | **Unlimited** | Node 24 + Python 3.14 + Astral `uvx` | **Dual MCP**: Stdio + Remote | Local File / In-Memory / SQLite | **$0** |
| **#1 Cloud** | **Regional Co-Located** | **Neon Functions (BYON)** *(Live)* | **1,000,000** (1M) | Linux Node.js (**2 vCPU, 8 GB RAM**) | Remote MCP (HTTP/SSE) | Serverless Postgres + `pgvector` + Cron | **$0** |
| **#1 Cloud** | **Regional Co-Located** | **Vercel Functions & Daemons** *(Live)* | **1,000,000** (1M) | Fluid Compute + Docker Daemons | Remote MCP (HTTP/SSE) | Edge Config / KV + `skills.sh` | **$0** |
| **#1 Cloud** | **Regional Co-Located** | **Supabase Edge (BYOS)** *(Live)* | **500,000** (500k) | Deno V8 Engine | Remote MCP (HTTP/SSE) | Postgres + `pgvector` + Realtime | **$0** |
| **#2 Edge** | **Worldwide Anycast Edge** | **Cloudflare Workers** *(Live)* | **3,000,000** (3M) | Global V8 Isolates (300+ cities) | Remote MCP (HTTP/SSE) | Workers KV & D1 Database | **$0** |
| **#2 Edge** | **Worldwide Anycast Edge** | **Azion Edge Functions** *(Live)* | **3,000,000** (3M) | Distributed V8 Isolates (500+ PoPs) | Remote MCP (HTTP/SSE) | Edge Storage & KV Caching | **$0** |
| **#2 Edge** | **Worldwide Anycast Edge** | **Fastly Compute (WASM)** *(Live)* | **10,000,000** (10M) | WebAssembly (Wasmtime/Lucet) | Remote MCP (HTTP/SSE) | Fastly KV & Secret Store | **$0** |
| **LIVE TOTAL** | **Regional Multi-Cloud Fleet** | **Neon + Vercel + Supabase** *(Live)* | **2,500,000 / mo** | **Live Serverless Fleet** | **Dual Stdio & Remote MCP** | **Postgres + pgvector** | **$0.00** |
| **FLEET TOTAL** | **Worldwide Multi-Cloud Fleet** | **All 6 Sovereign Clouds** | **18,500,000 / mo** | **Worldwide Edge Cloud Fleet** | **Dual Stdio & Remote MCP** | **Sovereign Multi-Database** | **$0.00** |

---

### Deployment Comparison: Self-Host vs. Regional Cloud vs. Worldwide Edge

| Capability | Local Self-Host (Docker / CLI) | #1 Regional Cloud (Neon, Vercel, Supabase · *Live*) | #2 Worldwide Edge (Cloudflare, Azion, Fastly · *Live*) |
|---|---|---|---|
| **Environment** | Local machine, VPS, Docker, Kubernetes | AWS Regional Zone (co-located with DB) | 300+ to 500+ Global Anycast Edge PoPs |
| **Tool Transport [T]** | **Dual-MCP**: Stdio (`npx`/`uvx`) + Remote HTTP/SSE | **Remote MCP** (HTTP/SSE) | **Remote MCP** (HTTP/SSE) |
| **Native Memory [I]** | In-Memory / File / Local SQLite / Postgres | Serverless Postgres + `pgvector` + Scheduled Cron | Edge KV / Fastly KV / D1 SQLite |
| **Cost** | **$0** (Free & Open Source forever) | **$0** (Runs on your provider's Free Tier) | **$0** (Runs on your provider's Free Tier) |
| **Account Ownership** | 100% Sovereign (Your Hardware) | 100% BYOC (Your Own Cloud Account) | 100% BYOC (Your Own Cloud Account) |
| **Quick Start** | `docker compose up -d` | `liate link` / `liate link --neon` / `liate link --supabase` | `liate link --cloudflare` / `liate link --azion` / `liate link --fastly` |

---

### Invoke from Any App or Webhook via HTTP (LAPI):

#### 1. Universal cURL

```bash
curl -X POST https://your-engine-endpoint/run \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Hello agent!",
    "L": "sarvam/sarvam-105b",
    "A": { "name": "sovereign_agent" }
  }'
```

#### 2. Node.js & Edge Fetch (`node -e`)

```javascript
node -e "
const url = 'https://your-project.supabase.co/functions/v1/openliate-engine/run';

const payload = {
  prompt: 'Check and list my inboxes using AgentMail tools.',
  sessionId: 'supabase-agentmail-session-1',
  L: 'sarvam/sarvam-105b',
  I: {
    memory: 'sessions/email-agent'
  },
  A: {
    name: 'email_agent',
    version: '0.0.1',
    intent: 'Autonomous Email Management Agent',
    maxSteps: 5
  },
  T: [
    'mcp:agentmail:list_inboxes,get_inbox,search_messages'
  ],
  E: {
    SARVAM_API_KEY: '\$SARVAM_API_KEY',
    AGENTMAIL_API_KEY: '\$AGENTMAIL_API_KEY'
  },
  keys: {
    SARVAM_API_KEY: process.env.SARVAM_API_KEY,
    AGENTMAIL_API_KEY: process.env.AGENTMAIL_API_KEY
  }
};

console.log('Invoking 5-line email-agent on Edge URL:', url);

fetch(url, {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json'
  },
  body: JSON.stringify(payload)
}).then(async res => {
  console.log('STATUS:', res.status);
  const text = await res.text();
  try {
    console.log('RESULT:\n', JSON.stringify(JSON.parse(text), null, 2));
  } catch {
    console.log('RAW BODY:\n', text);
  }
}).catch(console.error);
"
```

---

## [2] Benchmark of Liate Runtime Engine

Real empirical benchmarks measured live on `v0.1.0` standalone runtime engine against Sarvam 105B:

| Benchmark Metric | Measured Result | Environment & Notes |
|:---|:---|:---|
| **Hono Engine Cold Boot** | **`0.46 ms`** | Instant isolate startup |
| **LAPI Request Dispatch** | **`21.65 ms`** | `/run` JSON parsing & routing |
| **Sovereign LLM Latency** | **`3,866 ms`** | Sarvam 105B Reasoning (`sarvam-105b`) |
| **Token Generation Speed** | **`130.9 tok/s`** | 549 total tokens generated |
| **Full Pipeline Execution** | **`2,358 ms`** | Complete L · I · A · T · E ReAct turn |
| **Process Memory Footprint** | **`61.61 MB RSS`** | Extremely lightweight edge footprint |

```bash
# Run the benchmark yourself:
bun run benchmark
```

---

## <a id="aum-architecture"></a>[3] Architecture: The 5 Layers (AUM)

Every agent implements the **Agent Unified Manifest (AUM)** across 5 decoupled pillars declared as class properties or manifest keys:

| Layer | Name | Core (Built-in) | Community PRs |
|---|---|---|---|
| **L** | **LLM** | Sarvam AI Token Factory (2026 models) <sup>IN</sup> (`hub/llm/sarvam.json`), Neon AI Gateway (63+ models) (`hub/llm/neon.json`) | Add Claude, OpenAI, DeepSeek, custom OpenAI-compatible gateways... |
| **I** | **Integration** | Libraries (Zod, Math), in-memory state, persistent memory, Neon Postgres, Git repos | Add LlamaIndex, Redis, custom webhooks, vector stores... |
| **A** | **Agent** | Name, intent, ReAct loop, 1.26M+ `skills.sh` dynamic resolution | Add subagent swarms, custom multi-agent coordinators... |
| **T** | **Tools** | MCP v2 (Local stdio & Remote SSE), fetch, custom tools | Add filesystem, GitHub, custom MCP servers... |
| **E** | **Environment** | Pure `.env` key reader & execution configuration | Add custom secrets managers, edge isolate configs... |

> **Unifying the Two Greatest Standards:** Liate natively unifies **Vercel Labs' `skills.sh`** (1.26M+ portable domain skills injected into Layer **[A]**) and **Anthropic's Model Context Protocol (MCP)** (modular tools and actuators in Layer **[T]**). Zero custom glue code required.

---

## [4] License

[MIT](LICENSE) © 2026 OpenLiate. Free and open source forever.

<br>

<div align="center">

Built from India to the World

</div>
