Agent-Native is Builder.io's open-source framework for building applications where agents are first-class citizens, not bolted-on chatbots. The core primitive is a shared action: define work once and expose it to UI, agent, HTTP, MCP, A2A, and CLI simultaneously. Built on TypeScript with Drizzle ORM and Nitro server, it ships with an agent runtime that bundles chat, tools, skills, memory, jobs, observability, and handoffs into a single coherent stack. The framework includes production-ready templates — full SaaS apps (screen recording, design prototyping, content editing, slide generation) that demonstrate the agent-native pattern: cloneable, not scaffolded, with every line of code owned by the developer. At 1,900+ GitHub stars, it targets teams building product-grade agentic software who want to bring their own database, hosting provider, and model stack.
Agent-Native reframes how agents fit into applications. Instead of building a chat interface that calls tools, it provides a framework where agents and humans share the same actions, state, and UI surfaces. The key insight: if an action is worth exposing to a user, it’s worth exposing to an agent — and vice versa. The defineAction primitive enforces this by requiring a single implementation that serves all interaction modes.
The framework ships from Builder.io, the team behind Visual Copilot and a long line of developer tools. Their bet is that the next generation of SaaS applications will be agent-native from the ground up, not chat-augmented after the fact.
Agent-Native is built on three layers:
Actions are the atomic unit of work. Each action has a Zod schema, a run function, and automatic exposure to every interaction surface — web UI, agent tool calls, REST API, MCP server, A2A protocol, and CLI. This eliminates the typical pattern of maintaining separate handlers for human and agent interactions.
Agent Runtime bundles the agentic stack: conversational chat, tool calling, skills (composable action sequences), persistent memory, background jobs, observability hooks, and agent-to-agent handoffs. The runtime is opinionated about structure but flexible about models — bring any LLM provider.
Infrastructure uses Drizzle ORM for SQL-backed state (PostgreSQL, MySQL, SQLite) and Nitro for the server layer, enabling deployment to Vercel, Cloudflare, AWS, or any Node.js host.
The framework includes full-featured templates that serve as both examples and production starting points:
Each template is a complete, deployable SaaS application — not scaffolding.
Agent-Native targets teams building new applications where agents are a core feature, not an add-on. It’s a strong fit when you need shared actions across human and agent interfaces, SQL-backed persistent state, and a TypeScript-first development experience. For teams adding agent capabilities to existing applications, the framework may require more restructuring than alternatives that bolt onto existing codebases.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.
Open-source AI pair programming tool that works in your terminal to edit code across your entire repository.
AWS's AI-powered coding assistant that helps developers build, deploy, and optimize applications on AWS with code generation and transformation.