streaming
real-time token streaming with useAgentChat(). no manual SSE parsing. state managed for you.
react components
drop in an agent.
ship it.
chat, A2UI widgets, tool approvals. shadcn/ui compatible. connect to any inference shell agent.
pnpm dlx shadcn@latest registry add @inferenceshagent
how can I help?
ask me anything
blocks
everything your agent needs.
agent chat
full agent chat with state management, tool lifecycle, and A2UI widgets. streaming, approvals, widgets — all wired to a runtime. 4 lines to ship.
explore →agent.tsxtsx
import { Agent } from '@inference/common/agent' <Agent proxyUrl="/api/inference/proxy" config={{ agent: 'your-team/your-agent' }} />
A2UI widgets
A2UI-native agent widgets. flat component list in, rich UI out. spec-compliant, streaming-friendly, works on Gemini Enterprise.
explore →departureSFOsan francisco
✈️5h 45mon time
arrivalJFKnew york
san franciscopartly cloudy68°
⛅H: 72° L: 58°
today's tasks
your order
brown sugar lattelarge • oat milk • less ice
$6.50taro milk teamedium • regular sugar
$5.75total$13.32
tool invocation
pending, in-progress, completed, failed, awaiting approval. every tool state handled inline with collapsible details, approval UI, and widget results.
explore →arguments:
{
"query": "inference shell pricing"
}result:
{"urls":["https://inference.sh/pricing"],"snippets":["pay per second, no minimums"]}delete_file awaiting approval
arguments:
{
"path": "/data/old-backup.tar.gz"
}Q3 Revenue
generated bar chart with 3 data pointschart ready
chat primitives
container, messages, input, status indicators. composable building blocks for any chat UI. auto-resizing input, file upload, virtualized rendering for 1000+ messages.
explore →custom-chat.tsxtsx
import { ChatContainer, ChatMessages, ChatInput } from '@inference/common/agent' import { AgentChatProvider, useAgentChat } from '@inferencesh/sdk/agent' function MyChat() { const { messages } = useAgentChat() return ( <ChatContainer> <ChatMessages messages={messages} /> <ChatInput /> </ChatContainer> ) }
task output
streaming output, status indicators, tabbed views, auto-scrolling logs. display long-running agent task results with real-time updates.
explore →markdown
high-performance markdown renderer with syntax-highlighted code blocks, tables, images, YouTube embeds, and zoomable images. optimized for token-by-token streaming.
explore →getting started
build AI-powered apps with inference shell. the platform handles infrastructure, you write the logic.
typescript
const agent = new Agent({ tools: [searchWeb, analyzeData],})await agent.run("find recent papers on RLHF")agents can use tools, render widgets, and stream responses in real-time.
| feature | status |
|---|---|
| streaming | supported |
| tool calls | supported |
| widgets | 30+ types |
why inference shell ui
not just chat bubbles.
tool approvals
HITL gates render as UI. user clicks approve or reject inline. no modal interruptions.
A2UI widgets
agents render structured widgets — forms, cards, charts — not just text. typed with JSON schemas.
Next.js · Remix · SvelteKit · Hono · Express
no. it's a full A2UI widgets system. your agent can render tables, charts, forms, cards, not just text. components are typed with JSON schemas.
ready to ship?
start with the hosted platform. deploy your own when you're ready.
we use cookies
we use cookies to ensure you get the best experience on our website. for more information on how we use cookies, please see our cookie policy.
by clicking "accept", you agree to our use of cookies.
learn more.