# What is Aerostack?

> Aerostack is a developer infrastructure platform for building AI-native backends on Cloudflare's edge.

Aerostack is a developer infrastructure platform for building AI-native backends on Cloudflare's edge. It gives you **fullstack edge functions**, **MCP server orchestration**, and **intelligent bots** — all backed by built-in database, cache, queue, AI, vector search, and storage primitives.

  Aerostack is currently in **beta**. You're building on production infrastructure, but the API surface may evolve as we gather feedback.

## Agent Automation — One MCP, Any Agent

New to Aerostack? The fastest way to get started is to **copy one MCP config into your AI agent** and let it build everything for you. Your agent gets 16 tools to create functions, deploy MCPs, build bots, compose workspaces, and wire up workflows — all in one conversation.

**[Get started with Agent Automation →](/agent-automation)**

## The Problem

Building an AI backend today means stitching together 5-10 services: a database, a cache layer, a queue, an LLM provider, a vector database, file storage, auth — each with its own SDK, its own deployment, its own billing. Then you need to expose all of this as tools for AI agents to use.

## What Aerostack Does

Aerostack collapses this into one platform:

### 1. Fullstack Edge Functions

Write a function. It has native access to **everything** — database, cache, queue, AI, vector search, storage. Not HTTP calls to external services. Direct bindings with zero network latency.

```typescript

  async fetch(request, env) {
    // Native DB — no API call, direct binding
    const orders = await env.DB.prepare(
      'SELECT * FROM orders WHERE user_id = ?'
    ).bind(userId).all()

    // Native Cache — same datacenter, ~0ms
    await env.CACHE.put('recent-orders', JSON.stringify(orders), { expirationTtl: 3600 })

    // Native Queue — fire and forget
    await env.QUEUE.send({ type: 'order-notification', data: orders })

    // Native AI — multi-provider, built in
    const summary = await env.AI.run('@cf/meta/llama-3-8b-instruct', {
      messages: [{ role: 'user', content: `Summarize: ${JSON.stringify(orders)}` }]
    })

    return Response.json({ orders, summary })
  }
}
```

These functions power everything else in Aerostack — MCP tools, skills, bot logic, Agent Endpoints, and (coming soon) AI Proxy.

### 2. MCP Server Orchestration

Bring your AI tools together. Three ways:

- **Host on Aerostack** — Build an MCP server, deploy to edge
- **Proxy your existing MCP** — Register your server's URL, Aerostack becomes the gateway. Your API keys stay encrypted in Aerostack — team members never see them
- **Install from Hub** — One-click install community MCP servers (GitHub, Slack, Stripe, etc.)

Compose multiple servers into a **Workspace** — a single gateway URL that any LLM client can connect to.

```
Claude Desktop / Cursor / Your App
        |
   Workspace Gateway (single URL)
    |         |         |
Stripe MCP  GitHub MCP  Your Custom
(proxy)     (hosted)    Function
```

**For teams:** Each member gets a workspace token. Secrets stay in your vault. You see per-user analytics — who called which tool, when, how many times. Revoke one person without rotating production keys.

### 3. Intelligent Bots

AI bots that don't just chat — they **do things**. Connect a bot to a workspace and it can use every tool in it.

Two modes:
- **Agent Loop** — LLM autonomously decides which tools to call. Best for open-ended conversations.
- **Visual Workflow** — You design the exact flow. Built-in identity verification (`auth_gate`), conditional logic, code execution, human approval handoffs, scheduled messages, and bot-to-bot delegation.

Deploy to Telegram, Discord, WhatsApp, Slack, or any custom webhook — in minutes, not weeks.

### 4. One SDK

Install one package, get everything:

```bash
npm install @aerostack/sdk-react
```

Auth (email, OTP, magic links), Database (SQL + Postgres), Realtime (WebSocket pub/sub, presence, DB change events, message history), Storage (object storage with CDN), AI (multi-provider chat, embeddings), Cache (key-value with TTL), Queue (background jobs), Vector Search — all through React hooks or direct client calls.

Available for **React**, **Node.js**, **Go**, **Python**, and **Flutter**.

## Who is Aerostack For?

- **Developers building AI-powered products** — bots, agents, AI-augmented workflows
- **Teams using MCP** — centralize tool access, manage secrets, monitor usage
- **Startups shipping fast** — one platform instead of 10 services
- **Enterprise teams** — zero-trust MCP gateway with per-user audit trails

## What's in Beta

| Available Now | Coming Soon |
|---|---|
| **Agent Automation (MCP for agents)** | AI Proxy (smart API gateway) |
| Functions (fullstack edge) | |
| MCP Servers (host/proxy/hub) | |
| Skills (single-purpose tools) | |
| Workspaces (gateway composition) | |
| Bots (5 platforms + workflows) | |
| Agent Endpoints (REST API agents) | |
| SDK (React/Node/Go/Python/Flutter) | |
| All platform primitives | |

## Next Steps

- **[Agent Automation](/agent-automation)** — Copy one MCP config, let your AI agent build everything
- **[Quick Start](/getting-started/quick-start)** — Get your first project running in 3 minutes
- **[Architecture Overview](/getting-started/architecture)** — Understand how the layers connect
- **[Functions](/functions)** — Write your first fullstack edge function
- **[Bots](/bots)** — Build and deploy an intelligent bot