BuilderIO/agent-native

What it solves

Agent-Native addresses the gap between autonomous AI agents and user interfaces. Instead of limiting agents to a text-based chat box, it provides a framework where agents and UIs share the same capabilities, data, and state, allowing users to inspect, edit, and approve an agent's work through a purpose-built interface.

How it works

The framework operates on a shared layer of "actions"—functions defined once that can be called as tools by the AI agent or as code by the UI. This ensures consistent validation and permissions across both paths. It further synchronizes the experience through:

• Shared Data: Work performed by the agent is immediately visible in the UI, and vice versa.
• Shared State: The agent is aware of the current UI context, such as the active page or selected record.
• Action Layer: The agent interacts with the system via the action layer rather than simulating UI clicks.

Who it’s for

Developers building agentic applications in TypeScript who want to provide a rich, interactive user experience beyond simple chat interfaces.

Highlights

• Unified Action System: Define capabilities once for use by both LLMs and frontend code.
• Integrated Agent Surfaces: Includes built-in agent chat and support for agent teams (specialist agents).
• Persistent Context: Provides skills, memory, and a PostgreSQL backend for production (PGlite for local dev).
• Automation Support: Ability to run agent tasks on schedules or specific events.
• Extensible Infrastructure: Compatible with various LLMs, SQL databases, and Nitro-compatible hosts.