Agent-Native is an open-source TypeScript framework for building agentic apps: AI agents that come with their own rich UI. The agent and the UI share the same actions, data, and state, so anything you can do in the app, the agent can do too. Define an action once and it powers your UI, agent, HTTP API, CLI, MCP server, and A2A endpoint. Auth, permissions, and real-time collaboration are built in, and it works with any AI model. Build locally and deploy anywhere.

Hey Product Hunt 👋 Steve here, from Builder.io.

We've always felt that text is great for agents, but people need an interface for navigation, visuals, controls, and collaboration.

The goal is the flexibility of handing work to an agent, with the control of an app.

So we built Agent-Native, an open-source TypeScript framework for building AI agents with intuitive UIs. Anything you can do in the UI, the agent can do too.

It's for anyone who wants to take a workflow they know well and build an AI agent for it, with a UI that people and the agent both work in.

Actions are the core primitive. Define one, and it's available in the UI, agent, HTTP API, CLI, MCP, and A2A
Connect tools like Notion, Linear, and GitHub over MCP, and hand tasks to other agents over A2A
Shared instructions, skills, and memory, so the agent learns how your team works
Built-in auth, permissions, and real-time collaboration
Use any AI model, from Anthropic, OpenAI, and Gemini to local models with Ollama

Give it a spin:

npx @agent-native/core@latest create my-app

Agent-Native is in beta today, and we'd really value your honest feedback.

@steve_sewell Love the shared actions approach, define once and make it available across the UI, agent, and APIs.

Curious how permissions work when an agent takes actions on a user's behalf.

Wow, I was looking for something like this! I am learning a new language, and it's okay to see and interact with the dull ChatGPT UI, but I can just use this to build the UI that the agent can see and make the experience so much more interactive. Can see a ton more use cases! Congratulations to the team! 🎉