Is assistant-ui free?

A component library, not a hosted app

Most entries in this catalog are applications: clone them, run them, and you have a working product. assistant-ui is catalogued differently — as a component library built in the shadcn style, meaning the pieces are meant to be copied into your own codebase and owned from there, not installed as a black-box dependency you never open. That distinction matters more than it sounds: there's no account to create, no server assistant-ui runs on your behalf, and no usage limit to hit, because nothing about the library itself is a service.

The trade is the one every library makes against every application: you get flexibility and no vendor lock-in on the UI layer, at the cost of having to assemble the rest — a backend route, a model provider, and whatever authentication or persistence your product needs — yourself.

What's actually inside it

The primitives cover the parts of a chat interface that are fiddly to get right by hand: message lists that need to stay smooth as history grows, threads that track which message replies to which, and streaming responses that have to render token-by-token without the UI jumping or flickering. Tool calls — the model asking to run a function and the interface showing that request and its result — are a newer, messier problem than plain chat text, and generative UI extends that idea further: letting a model's response render as an actual interface component instead of a wall of text.

The two named integrations point at how the library expects to be wired up. The Vercel AI SDK is the layer most commonly used to talk to a model provider and stream its response back to the browser, handling the swap between OpenAI, Anthropic or another provider as configuration rather than a rewrite. LangGraph is built for a different job: orchestrating multi-step, stateful agent behaviour rather than a single request-response turn. assistant-ui integrating with both means it can sit on top of a simple streaming chat or a more elaborate agent, without you rebuilding the interface for either.

What you still have to bring

Because it's a UI layer, assistant-ui doesn't include a model provider, an API key, or a server to call. You need a backend endpoint — typically an AI SDK route or a LangGraph deployment — behind it before any of the primitives have something to render.

It also doesn't include authentication, chat history persistence, or multi-model switching as a finished feature. Those are things you build using the primitives, not things that arrive pre-wired. If that sounds like more assembly than you want, the next section is for you.

If you want a full app instead of a library

Vercel AI Chatbot is Vercel's own open-source reference implementation: a Next.js App Router app built on the AI SDK, with a streaming chat interface, model-provider switching across OpenAI, Anthropic, xAI and others, authentication, and chat history persistence already wired together. Because the AI SDK team maintains it directly, it tends to track that SDK's current best practices closely.

Chatbot UI, built by Mckay Wrigley, is one of the most-forked AI chat codebases on GitHub: multi-model chat across OpenAI, Anthropic and local models through Ollama, with folders and saved prompts, file attachments, and Supabase-backed persistence for accounts and history.

Morphic takes a different shape again — a Perplexity-style AI answer engine rather than a plain chatbot, built on Next.js and the Vercel AI SDK, with pluggable search providers (Tavily, SearXNG or Exa), model switching, and chat history. If what you actually want is a search-and-cite experience rather than a conversation window, this is the closer match.

Picking between assistant-ui and a full template

The decision usually comes down to how much of the product already exists. If you're adding a chat feature to a SaaS that already has its own auth, database and routing decided, assistant-ui slots the interface in without dragging along someone else's opinions on any of those — you keep what you have and only borrow the chat primitives. If you're starting from nothing and want a working chatbot today, one of the three ready-made applications above gets you further, faster, because the account system and history are already solved rather than left for you to wire.

The catalog's AI Apps category holds nine entries in total, assistant-ui among them — the rest split between full chat applications like the three above, an AI answer engine, a code-interpreter sandbox template, and an AI-agent scaffold, worth browsing if none of the three chat apps above is quite the shape you need.

Frequently asked