Now in public beta

Pre-built React components
for AI agent UI

Copy, paste, own the code. Six production-ready primitives for progress tracking, tool call streams, human-in-the-loop approval, plan visualization, token metering, and trace debugging — all wired to AG-UI out of the box.

Browse components → GitHub
AG-UI native React + TypeScript Copy-paste, not npm Framework-agnostic Open source

Everything an agent UI needs.
Nothing it doesn't.

Fetch context1.2s
Analyze documents3.7s
Generate response

ProgressTracker

Multi-step tracker for long-running agent tasks. Live duration tickers, ETA, failure and skip states.

Live demo →
bash running
$ git diff --staged
+ const handleApproval = () => {
+ dispatch({ type: 'APPROVE' })

ToolCallCard

Collapsible card with tool name, params, and streamed output. Handles code diffs, JSON, and markdown.

Live demo →
🔐 Approval required
Delete 47 records from users where inactive > 90d
HIGH RISK · Irreversible

HumanApprovalModal

Interrupt-driven modal that blocks the agent until a human approves, edits, or rejects the proposed action.

Live demo →
Research phase
Fetch sources
Summarize
Synthesize

PlanVisualizer

Hierarchical tree of the agent's execution plan. Streams live as nodes are added during reasoning.

Live demo →
Tokens used 24,318
Est. cost $0.34
Budget cap $1.00

TokenMeter

Running token count and cost tracker. Budget cap with green/amber/red/locked states per provider.

Live demo →
00:00.12 tool search("AI frameworks")
00:01.08 llm Chose top 3 results…
00:01.94 tool fetch(url[0])
00:03.21 event TEXT_MESSAGE_CONTENT…

TraceViewer

Step-by-step execution trace with timestamps, token counts, and collapsible decision tree for debugging.

Live demo →

The AI SDK solved the backend.
Nobody solved the frontend.

Progress bars that don't lie

Hand-rolled progress UIs either spin indefinitely or jump from 10% to 100%. ProgressTracker streams real step completions with live duration tickers — so the bar actually reflects what the agent is doing.

Approval modals that feel inevitable

Most agents skip human-in-the-loop entirely because building the modal is painful. HumanApprovalModal handles the interrupt, the diff preview, the keyboard shortcuts, and the AG-UI event wiring — in one copy-paste.

Trace viewers that make debugging obvious

When an agent does something unexpected, you need to see every LLM call, every tool invocation, every token — in order, with timestamps. TraceViewer surfaces that without you writing a single log formatter.

Components you actually own

No npm dependency that breaks on a major version. No proprietary API. You copy the JSX into your project, modify it freely, and ship it. Modeled exactly on the shadcn/ui philosophy — because it's the right one.

Vercel AI SDK
CopilotKit
Assistant UI
TaskFrame
Progress tracker
Tool call cards
Partial
Approval modals
Plan visualizer
Token meter
Trace viewer
Copy-paste, own the code

Works with any framework that emits AG-UI events

AG-UI is an open protocol for streaming agent state — think of it as SSE with a typed event schema. If your backend emits AG-UI events, TaskFrame components wire up automatically. No adapter needed.

Already used by LangGraph, CrewAI, Mastra, Pydantic AI, and anything else that speaks the protocol. The event stream is framework-agnostic — if your agent backend can emit JSON over SSE, it works.

LangGraph CrewAI Mastra Pydantic AI CopilotKit + any SSE backend
ag-ui event stream
RUN_STARTED thread_id: "t-8f2k"
STEP_STARTED step: "fetch_context"
TOOL_CALL_START tool: "search"
TOOL_CALL_END duration: 342ms
STEP_FINISHED status: "success"
HUMAN_INPUT_REQUEST action: "approve_delete"

Copy. Paste. Own the code.

No npm dependency. No version lock. Download the JSX you need, drop it in your components/ folder, and ship. That's the shadcn/ui philosophy — and it's the right one for agent UIs.

Works with Next.js, Vite, Remix, or any React project. Tested with TypeScript. AG-UI event wiring means it plugs into your existing agent backend without glue code.

# 1. Download the component into your project
curl -O https://taskframe.polsia.app/ProgressTracker.jsx
mv ProgressTracker.jsx src/components/

# 2. Install peer dep (if not already installed)
npm install react

# 3. Import in your page
import ProgressTracker from './components/ProgressTracker'

Stop building agent UI from scratch.
Start with TaskFrame.

Open source. AG-UI native. Six copy-paste components for every state your agent will ever be in.