Agent Native

Overview

Agent Native is an open-source framework from BuilderIO for building "agent-native applications" — a new paradigm where AI agents don't just chat alongside an application but actively act inside it. Unlike traditional chatbot integrations, Agent Native provides a unified action system that simultaneously powers UI interactions, agent tool calls, API endpoints, MCP servers, A2A (Agent-to-Agent) communication, and CLI commands from a single definition.

The framework enables real-time multiplayer editing where humans and agents collaborate as peers on the same document, and agents can see exactly what the user is looking at for context-aware assistance. With over 755 releases as of June 2026, Agent Native is under rapid, active development.

Features

• Unified Action System: Define work once. Use it from UI, agent, API, MCP, A2A, and CLI simultaneously.
• Agent Runtime: Bundles chat, tools, skills, memory, jobs, observability, and handoffs into a cohesive runtime.
• Real-time Multiplayer: Humans and agents edit documents concurrently as peers with conflict resolution.
• Context Awareness: The agent sees what the user is looking at and supports text selection with Cmd+I invocation.
• Self-improving: Agents can add features, fix bugs, and refine the UI over time based on user feedback.
• A2A Protocol Support: Agents can tag and coordinate with each other across different applications.
• Backend Agnostic: Works with any Drizzle-supported SQL database and Nitro-compatible host.
• Ready Templates: Fully open-source SaaS app templates including Clips, Plans, Design, Content, Slides, and Analytics.

Installation

npx @agent-native/core@latest create my-app
cd my-app
pnpm install
pnpm dev

To add a skill (e.g., visual planning) to an existing setup:

npx @agent-native/core@latest skills add visual-plan

Quick Start

import { defineAction, agent } from '@agent-native/core';

// Define an action once
const sendEmail = defineAction({
  name: 'sendEmail',
  input: { to: 'string', subject: 'string', body: 'string' },
  handler: async ({ to, subject, body }) => {
    // Send email logic
    return { success: true };
  }
});

// The action is automatically available as:
// - A UI component (React)
// - An agent tool
// - An API endpoint (REST)
// - An MCP tool
// - An A2A capability

Core Concepts

Actions

Actions are the fundamental building block of Agent Native. Each action is defined once and automatically exposed across all surfaces — UI, agent, API, MCP, A2A, and CLI.

Agent Runtime

The runtime provides built-in support for chat, tool execution, reusable skills, persistent memory, background jobs, observability via OpenTelemetry, and agent-to-agent handoffs.

Multiplayer Collaboration

Agent Native treats agents as first-class collaborators in the editing process. Multiple agents and humans can work on the same document simultaneously, with real-time conflict resolution.

Advanced Features

• Skill System: Reusable packages of actions and agent instructions that can be composed together
• Agent Memory: Persistent memory across sessions using vector embeddings and SQL storage
• Background Jobs: Schedule and execute long-running agent tasks asynchronously
• Observability: Built-in OpenTelemetry tracing for all agent actions and LLM calls
• MCP Integration: Expose any action as an MCP tool for use by other AI agents
• A2A Communication: Cross-application agent coordination via the Agent-to-Agent protocol

Examples

• SaaS Platform: A content management system where agents can create, edit, and publish content alongside human authors
• Analytics Dashboard: An AI data analyst that can modify charts, add new visualizations, and respond to natural language queries
• Design Tool: A collaborative design environment where agents can manipulate UI components, adjust layouts, and suggest improvements
• Project Management: A planning tool where agents can create tasks, update statuses, and coordinate across team members

Pros

• ✅ True agent-native architecture — agents work inside apps, not alongside them
• ✅ Real-time human-agent collaboration as peers
• ✅ Unified action system dramatically reduces code duplication
• ✅ MIT licensed with fully customizable, forkable templates
• ✅ Self-improving agents that can modify the application
• ✅ Multi-protocol support (MCP, A2A, REST, CLI) from one definition

Cons

• ❌ Early stage with smaller community (3,000 stars)
• ❌ TypeScript/Node.js only — no Python support
• ❌ Documentation still evolving alongside rapid development
• ❌ Requires significant infrastructure (Node.js, database)
• ❌ Production readiness varies across features

When to Use

Use Agent Native when:

• You want to build a new SaaS application with AI agents as first-class citizens
• You need both a traditional UI and AI agent access to the same functionality
• You're building collaborative tools where humans and agents work together
• You want to expose your application functionality via MCP or A2A protocols
• You prefer TypeScript and the Node.js ecosystem

Resources

• Official Website
• GitHub Repository
• NPM Package