The Assistant SDK for Android
The Assistant SDK puts an agent in your app that answers with your own components instead of text, rendered as native Jetpack Compose from the components and APIs you already have. Maven artifact under ai.metabind, install today.
Every answer renders as native Jetpack Compose from a single BindJS definition, with Material 3 defaults, native scrolling, animation and accessibility, at 60 FPS.
Font scaling, TalkBack and accessibility behave like the rest of your app, because it is the rest of your app.
The same components, schema, validation and allowlist in your iOS, Android and web apps, and in Claude and ChatGPT. The agent can't render on iOS what you didn't approve on Android.
Scan a code in Metabind Studio to see any component natively on Android through the Metabind app. An iPhone opens it as an App Clip, with nothing to install, and the web in your browser.
Your agent can show only the components you approved, each validated against its schema before it renders, the same way on every platform. The Agent Proxy runs it with your system prompt and your own model key.
The Agent Proxy runs your agent on Anthropic, OpenAI or Gemini. Your own model key is encrypted server-side, so it never ships in your app, and on paid plans every turn can go to your own Langfuse project.
Tool calls, streaming responses and schema validation on every render. The SDK implements the spec and keeps up with it, so your app wires up a view instead of a protocol.
Greet like this. Ask X before Y. Never leave the catalog. Write it as a system prompt in Metabind Studio, and the Agent Proxy sends it to the model on every turn.
Retail and finance: two complete apps, each with the agent inside. Clone one, point it at your APIs and run the whole agent inside your own app.
An interior design assistant for a home furnishings catalog. Product carousels, side-by-side comparisons, color palettes and generated room inspiration all come back as rendered components, so the conversation stays visual instead of collapsing into a wall of text.
View and Clone
A personal finance app built entirely from the agent's own tools. Instead of a chat window, they render as native tabs and sheets, so the app you tap and the agent you ask share one definition.
View and Clone
No. Your AI coding agent converts the components you have into BindJS, from SwiftUI, Jetpack Compose, React or a Figma file, and the Metabind CLI imports them. Your APIs become Data tools, and you write the agent's system prompt in Metabind Studio. We generate and host the MCP server; you keep your app. You can ship on the Component Kit, in your brand, before conversion is done.
Yes. Add your project's MCP endpoint beside the servers your agent already uses, and use only the SDK's renderers: they draw its answers natively, with the same components and the same validation.
The SDK is Apache 2.0 and open on GitHub. The agent it embeds runs on the platform: you define the tools and write the system prompt in Metabind Studio, and we generate and host the MCP server. Tool definitions and schemas are exportable, and the protocol is standard MCP, so nothing proprietary sits in the client path.
The Free plan includes three MCP Apps, 10K tool invocations, Metabind Studio, BindJS and the Assistant SDK. No credit card.
No credit card. No sales call. No server to run.