CopilotKit vs assistant-ui vs the Vercel AI SDK
A side-by-side comparison of CopilotKit, assistant-ui and the Vercel AI SDK, from each project's own pages.
Quick answer
Use the Vercel AI SDK when you want the most control: it is an Apache-2.0 TypeScript toolkit for model calls, tools and chat hooks, and you build the interface. Use assistant-ui when you want polished chat components on top of whatever backend you have: it is MIT, with an optional paid cloud for thread history. Use CopilotKit when the agent already exists and you want it in front of users with shared state, generative UI and human-in-the-loop: it is MIT, with a paid hosted platform. All three expect your team to build and run the agent.
These three come up in almost every conversation about putting AI chat into a React app. They overlap less than it seems: one is a toolkit for the whole AI layer, one is a component library for the chat, and one is a stack for connecting an agent you already have to the people using it. Many teams use two of them together.
Every claim below comes from each project's own pages, read on 20 November 2026.
Side by side
| CopilotKit | assistant-ui | Vercel AI SDK | |
|---|---|---|---|
| In its own words | The open-source stack for putting AI agents in front of real users | A library to build production-grade AI chat experiences fast | The TypeScript toolkit for building AI applications and agents |
| Licence | MIT | MIT | Apache-2.0 |
| Where it runs | React, Angular, Vue, React Native, Slack, Microsoft Teams | React, plus React Native, Vue and the terminal | Hooks for React, Svelte and Vue; any JavaScript runtime |
| Agent backend | The agent you already built, on any framework, over its AG-UI protocol | AI SDK, LangGraph, LangChain, AG-UI, A2A, your own stream | It is the backend toolkit: model calls, tools, structured output |
| Approvals in the UI | Human-in-the-loop | Inline human approvals | You build them |
| Hosted option and price | Free hosted tier of 1,000 interactions a month; $49 a month for the first 10,000; custom annual | Optional cloud: free to 200 MAU; $50 a month for 500 plus $0.10 each after; custom Enterprise | None needed; you pay your model provider |
The Vercel AI SDK: the foundation
The AI SDK sits lowest. It gives you one way to call models across providers, define tools, get structured output and stream it, plus UI hooks such as useChat. It does not give you a finished chat interface; you build that, or put a component library on top. Vercel's own component library for it, AI Elements, says it will be deprecated at the end of November 2026 as it moves to shadcn/ui.
Pick it when you want full control over every layer and are happy to write the interface.
assistant-ui: the chat interface
assistant-ui is the chat. It gives you the thread, messages, composer, attachments, tool calls rendered as your own React components and inline approvals, as primitives you can style. It runs on top of whichever backend you choose, and lists the AI SDK first among them. Thread history can live in your own storage or in its optional cloud.
Pick it when you want a polished chat quickly, on a backend you already have or plan to build with the AI SDK.
CopilotKit: the agent in front of users
CopilotKit starts from the other end: you already have an agent, built with LangGraph, CrewAI, Mastra, Google ADK or anything else, and you want it inside your app. It connects over AG-UI, an open protocol it maintains, and adds shared state between the agent and the app, generative UI and human-in-the-loop, across more platforms than the other two.
Pick it when the agent exists and should read and change what is on screen, not only reply in a panel.
And if you do not want to build the agent
I build Verb, so weigh this accordingly. Verb is an AI agent that sits inside your product and does the work your users ask for, as the signed-in user, through your existing API. You describe what it may do, try it on a test copy, and add one script tag; it is integrated in under an hour in most cases. It is the wrong choice if the agent is your product and you want to own every line, and one of the three above is the right one.
For more options, see CopilotKit alternatives and assistant-ui alternatives.
Sources
Each read on 20 November 2026.
Frequently asked questions
Can I use assistant-ui with the Vercel AI SDK?
Yes. assistant-ui lists the Vercel AI SDK first among its supported backends, with its own @assistant-ui/ai-sdk package. Many teams use the AI SDK for the model and tools and assistant-ui for the interface.
Which is best for an agent built with LangGraph?
CopilotKit and assistant-ui both support LangGraph. CopilotKit lists it among the agent frameworks it connects to over its AG-UI protocol; assistant-ui has LangGraph and LangChain packages.
Which one is free?
All three libraries are free and open source: CopilotKit and assistant-ui are MIT, the AI SDK is Apache-2.0. The paid parts are optional hosted services: CopilotKit's platform from $49 a month after a free hosted tier, and assistant-ui's cloud from $50 a month after 200 free monthly active users (as of 20 November 2026).
Is there an option that does not require building the agent?
Verb. The three here are for teams building their own agent. Verb is an agent you set up inside your product: it does the work users ask for as the signed-in user, through your existing API.

Vishnu, founder of AskVerb @itsvishnups
Building the AI agent that lives inside a SaaS product and does the work its users ask for. Writing down what I learn along the way, including the parts I got wrong first.