Open-source React components for AI agent interfaces: conversations, tool calls, approvals, background runs, and rich outputs.
A chronological stream for reasoning, searches, terminal work, and tool execution.
A compact task summary with the current activity and an optional task tray.
A human-in-the-loop decision surface for questions, review steps, and consequential actions.
File and image attachments with progress, failure recovery, metadata, and removal actions.
A controlled playback surface for speech, generated audio, and recorded tool output.
Work still running elsewhere and completed results waiting to be collected.
Ongoing and completed agent jobs collected in one actionable inbox.
A dropdown with vertical category tabs on the left and their items on the right, with a sliding tab indicator, animated height, and a morphing chevron.
A compact source summary that expands into grounded references with useful context.
A muted code surface with line numbers, focused rows, filename metadata, and copy feedback.
A Tailwind palette picker with search, hex and opacity inputs, and optional linear, radial, and conic gradients with draggable stops.
An expandable changed-file summary with folder groups, line counts, and a diff action.
Clear connected, reconnecting, resumed, and offline states for resumable agent streams.
Where an agent context window went, with token attribution and remaining capacity.
Run cost split by model or task and compared against an optional session budget.
A virtualized, sortable table with resizable columns, a sticky header, and left or right pinned columns for scanning dozens of fields at once.
A source card that preserves document identity, location, metadata, and the cited excerpt.
A user message with copy and edit actions, inline editing with keyboard shortcuts, and switching between edited versions.
A minimal email draft with inline fields, a spacious message, and copy, expand, and send actions.
A secret-aware configuration surface with masked values and explicit reveal or copy actions.
Response feedback with copy, thumbs up and thumbs down actions that expands into a reason picker and an optional note.
A compact code-change surface with old and new line numbers plus live change counts.
A keyboard-navigable file tree with collapsible folders, git change badges, and change counts rolled up to folders.
Follow-up prompts that stagger in after an answer, with number key shortcuts and a selected state.
A keyboard-driven next step picker with numbered options and a free text custom question.
A streaming image surface with live progress, a playable Snake diversion, and an in-place completed result.
A numbered reference inside prose with an inspectable source preview.
A weighted multi-stage progress surface for work measured in minutes rather than seconds.
A horizontal link preview with an image, seller details, and optional product pricing and ratings.
Visible, removable facts an agent retains about the user or ongoing work.
A focused conversational surface with distinct user, assistant, and system treatments.
Pending user turns that remain visible, removable, and immediately sendable while an agent is busy.
A compact conversation minimap that navigates user turns while keeping the full thread visible.
A compact model menu with descriptions, plan badges, effort controls, and additional models.
Shared card shadow, surface shadow and heading/label utilities used across Pasta UI components.
A capability-level consent surface that explains reach and duration before access is granted.
A proposed plan with clear steps and controls to edit, cancel, or start the work.
An auto-growing agent composer with attachment, model, submit, and stop controls.
Usage remaining, reset timing, and an upgrade path without hiding the limit.
A compact dropdown for choosing how much reasoning power a response should use.
A timed reasoning stream that follows the latest thought, then folds into an inspectable duration summary.
A long-form report reader with a contents rail, reading progress, and sections that stream in as they are written.
Ranked source passages with a searching state, highlighted query terms, relevance bars, and expandable text.
A recurring agent job with human-readable timing, next run, and pause controls.
Ranked search results with site icons, titles, source URLs, and plain text category tabs.
One delegated task with state, timing, assigned agent, and returned result.
A quiet live status that keeps the agent current action readable while work continues.
A live agent plan with completion counts, checkable tasks, and details you can expand.