# Introduction

> What the Divinci AI SDK is, how the packages fit together, and which one to use.

The Divinci AI SDK is a family of TypeScript packages for building on the Divinci
platform — chat, retrieval-augmented generation (RAG), fine-tuning, evaluation,
and pay-per-call tools. Every package shares one type system (`@divinci-ai/types`)
and the same authentication model, so what you learn in one carries over to the rest.

## The platform in one paragraph

You configure an AI assistant once as a **Release** — its model, prompt, RAG
knowledge base, theme, and access rules. A Release lives inside a **Workspace**
(also called a white‑label). Your **end users** chat against a Release; your
**backend** manages Workspaces, Releases, and knowledge bases. The SDK packages
split cleanly along that line: client-side packages talk to a Release, the server
package administers the platform.

## Which package do I need?

<CardGrid>
  <Card title="@divinci-ai/client" icon="laptop">
    A headless, browser‑safe TypeScript SDK. You own the UI; the client owns
    chat threads, streaming, realtime WebSockets, and per‑user RAG. Reach for
    this when building a custom chat experience in React, Vue, Svelte, or
    vanilla JS.
  </Card>
  <Card title="Embed script" icon="rocket">
    A zero‑build `<script>` tag that mounts a fully‑rendered Divinci chat widget
    on any page. No bundler, no npm. Reach for this when you want chat live in
    minutes and don't need to control the UI.
  </Card>
  <Card title="@divinci-ai/server" icon="node">
    A Node.js SDK with full platform access — Workspaces, Releases, RAG ingestion,
    API keys, fine‑tuning, QA suites, and x402 payments. Holds your secret API key,
    so it runs only on a trusted backend.
  </Card>
  <Card title="@divinci-ai/mcp" icon="puzzle">
    A Model Context Protocol client. Connects MCP‑compatible assistants (Claude
    Desktop, Cursor, your own agent) to Divinci tools and resources, with optional
    x402 pay‑per‑call.
  </Card>
</CardGrid>

<Aside type="tip" title="Rule of thumb">
  If the code runs in a **browser**, use `@divinci-ai/client` (or the embed
  script). If it runs on a **server** with a secret API key, use
  `@divinci-ai/server`. If it bridges an **AI assistant** to Divinci tools, use
  `@divinci-ai/mcp`.
</Aside>

## How they fit together

A typical production app uses two packages at once:

- Your **backend** uses `@divinci-ai/server` to create a Release, upload RAG
  documents, and mint short‑lived user tokens.
- Your **frontend** uses `@divinci-ai/client` with that token to run the chat,
  streaming responses directly to the browser.

The `@divinci-ai/types` package is the shared vocabulary — `ChatMessage`,
`Release`, `RagDocument`, and friends. It ships as a dependency of the other
packages, so you rarely install it directly, but you can import its types for
your own function signatures.

## Next steps

<LinkCard title="Installation" href="/getting-started/installation" description="Install the package you need with npm, pnpm, or yarn." />
<LinkCard title="Quick Start" href="/getting-started/quickstart" description="Send your first message and run your first server operation." />
<LinkCard title="Authentication" href="/getting-started/authentication" description="API keys, user tokens, and external users — and where each belongs." />