An AI integration platform built for developers

In minutes,
give your website
an AI assistant.

Define a few tools and the platform generates the LLM schema and TypeScript types for you; embed one browser SDK snippet and AI can drive your page directly with natural language — filling forms, navigating, clicking buttons.

Open Console See a 3-step example ↓
No key needed for local dev · npm install to wire up the SDK · type-safe front-end code
Why onagent

Everything you need to wire AI into your UI

From tool definitions to type-safe front-end code and secure isolation, all in one place.

Define once, get both

One tool definition auto-generates the LLM's JSON schema and a type-safe TypeScript ToolHandlers interface — schema and front-end types never drift apart.

AI drives your UI directly

Tool calls dispatch straight to the handler you registered: fill forms, navigate, check boxes, click buttons — not just a reply, but a real action.

Per-app isolation & allowlisting

Every API key resolves to its app on the server (the front end can't spoof it) and is bound to a single Origin — unset means fail-closed by default.

Ship in 3 steps

Stand up an AI-powered feature in three steps

Define a capability, push it to the platform, embed it in your front end — click a step on the left to see the actual code for it.

↳ Once wired up, the AI picks a matching tool for each prompt — the SDK calls your handler and reports the result back to the model.

tools
appId: demo-app
tools:
  - name: fill_search_form
    description: Fill the on-page search box and submit it.
    parameters:
      type: object
      required: [query]
      properties:
        query:
          type: string
          description: The keyword to search for.
# Sign in (opens a browser to authorize; the URL only carries a one-time code)
$ atp login --web

# Push your tool definitions
$ atp save-tools demo-app tools
✓ Saved 1 tool to "demo-app"

# List your apps any time
$ atp list-apps
demo-app        1 tool, has key
import { AgentBridge } from "@onagent/agent-bridge-sdk";

const bridge = new AgentBridge({
  url: "wss://your-backend/ws",
  appId: "demo-app",
  apiKey: "YOUR_APP_API_KEY",
  tools: {
    fill_search_form: ({ query }) => setQuery(query),
  },
});

bridge.sendContext({ query, path });  // send page state as background context
bridge.prompt("search for red sneakers");     // let AI take action
What you can build

Drive real UI actions with a single sentence

These aren't mockups — they're runnable examples under examples/ in the repo.

examples/react-demo · React

Checkout assistant

Let AI fill in a search, navigate between pages, and even click the checkout button — every action maps to a handler you registered.

check me out → click_checkout_button()
Your order has been placed ✓
examples/analysis · Vue 3

Survey analysis assistant

On a list of survey questions, let AI select the relevant ones in natural language, ready to run statistical analysis.

select the questions about age → select_question(["p2q1", …])
Selected 3 questions

Ready to bring AI into your UI?

Create an app, issue an API key — that's it. Local development works with no LLM key at all.

Open Console Revisit the 3 steps