AI Agent UI Components for React

Components for the screens an AI agent works in: its plan, its tool calls, its reasoning, the answer as it streams and the moments it needs your approval. Built with Tailwind CSS and Motion, ready to install via the shadcn CLI.

Premium

Permission Request

A permission card where an agent asks for access and you allow it once, for this chat or always for a pattern, then revoke it from the receipt.

Premium

Agent Status

A status pill that says what the agent is doing, whose move it is, time and cost, and opens to the step history and token bill.

Premium

Code Block

Code an agent just wrote streams in line by line, then numbered markers tie its notes to the exact lines, with Copy and Apply.

Premium

Agent Plan

An agent's step plan you can edit and start, then watch run one step at a time with live activity and a segmented progress bar.

Premium

Diff Review

Review an agent's proposed code edits change by change: read the reason, keep or discard each one, then apply only what you kept.

Premium

Cited Answer

An answer with numbered citations: pick a number or a source to see the quoted passage and whether it really backs the sentence.

Premium

Agent Composer

The message input an agent sits behind: a textarea that grows to six rows, inline slash commands, @file mentions as removable chips, a drag-and-drop attachment row, a model picker with each provider's logo, and a round send button whose glyph becomes a pulsing stop square while the agent works.

Premium

Streaming Answer

An assistant answer that streams in word by word, already formatted, with new words fading in, a stop and continue control, and message actions.

Premium

Agent Activity

The audit trail of one agent run: every action on a rail of icons, one line each, ending in what it produced.

Premium

Thinking Trace

A single steady line shows what an agent is thinking right now, then settles into how long it thought, with every step one click away.

Premium

Checkpoint

An agent checkpoint card that pauses for a decision on a question, command or plan, then collapses into a one-line receipt with undo, edit-before-allow and a risk grade.