Most “AI in your app” bolts a chatbot onto a sidebar. agent-native, a new MIT-licensed framework from Builder.io, starts from the opposite premise: the agent and the interface are two views of the same application, so anything the UI can do, the agent can do — and vice versa.
## What agent-native does
The agent can see what’s on your screen, change it, and navigate you between pages, while you keep clicking buttons normally. You can invoke the same actions from code, CLI, HTTP, MCP, or A2A, so a workflow isn’t locked to one entry point. It ships a chat interface with native tables, charts, approvals, and setup flows, plus a full product UI where chat can start center-stage, slide to a sidebar, and stay synced with app state.
## One surface, many protocols
The pitch for builders is interoperability. agent-native plugs into A2A, MCP and remote MCP OAuth, the Claude Agent SDK, the Vercel AI SDK, OpenAI, and AG-UI through adapter runtimes. Instead of betting on one agent stack, you wire your app once and let it speak whichever protocol the moment demands.

Leave a comment