Shadcn AI Blocks: Chat, Copilot and Agent Activity UI
Most assistants stop at a bubble list and a text input. These 20 blocks are the rest: the composer, grounded answers, and a record of what the agent did.
Shadcn AI and Chat Blocks for Copilots and Assistants
An assistant is easy to demo and hard to ship. The demo is a transcript and a text box. The product is everything around it: what the first screen says before anyone has typed, which files the model is allowed to read, what a reply looks like when the answer is a table rather than a paragraph, what the reader does when the answer is wrong, and where the whole thing sits when the page underneath still has to be usable.
This group collects those surfaces. Full page copilots with thread switching and model pickers, panels docked beside a document that can write their reply into it, welcome screens whose starters actually load context, a split pane that runs one prompt against 2 models, a voice first panel, a branching thread that versions an edit instead of overwriting it, and a docs assistant that shows which articles it read before answering.
Every block is UI only. Streaming, tool traces, citations and run status render from typed local state, so the same surface fronts the Vercel AI SDK in one app, LangGraph in another, and a plain fetch loop in a third. Both twins ship, Base UI and Radix, across every ReUI style.
Browse Shadcn AI Block Categories
Three categories: the chat surfaces people talk to, the views that show what an agent did while it worked, and what that work costs.
- AI Chat13 blocks
Copilot shells, docked and floating assistant panels, welcome screens, model comparison, voice turns, branching threads and grounded docs support.
- Agent Activity5 blocks
Agent run views, from live step feeds and approvals to research traces, MCP call tables, and run reports.
- AI Usage2 blocks
Spend planning and month to date spend views for AI product workspaces, by token volume, model, budget and caps.
What These Blocks Handle That a Transcript Does Not
- The screen before the first message A chat that opens on nothing teaches nothing. The welcome screens carry live workspace stats, starters that fill the composer and attach their context, and resume rows for threads already in flight.
- Scope decided before the answer Connected sources toggle in the composer chrome, and a source switched off is named as a gap in the reply rather than quietly guessed around.
- Answers with a shape Replies render as small documents: sortable tables, folded reasoning, code artifacts, comparison bars, approval gates and citation chips, not one long paragraph.
- Somewhere to live Full page, a right sidebar that pushes the page, a floating panel, or a Sheet on mobile. The page underneath stays usable in all of them.
Chat Blocks Versus a Chat SDK
An SDK gives you the stream. It does not give you the stop control, the retry on a step that failed, the thread switcher over pinned and recent chats, the attachment chips you can remove before sending, or the empty state that runs before the first token arrives. That is the part teams rebuild every time, and it is the part that is here.
Nothing in this group imports an AI framework. Install a block, point its typed props at whatever produces your tokens, and keep your own transport.