agent-native: a framework for agent-native apps where agents and UI share the same actions, state, and context

What it solves

It eliminates the friction between an application's user interface and its AI agents. Instead of building separate logic for a UI button and an AI tool, developers can define a single action that works across all surfaces, including the UI, agents, HTTP, MCP, A2A, and CLI.

How it works

The framework uses an "action-first" approach where developers define a schema and a run function for a specific task. This single definition is then shared across the agent runtime and the UI. The runtime provides built-in support for chat, tools, skills, memory, jobs, and observability. It is backend-agnostic, allowing developers to plug in any Nitro-compatible host and any SQL database supported by Drizzle.

Who it’s for

Developers building "agent-native" applications where AI agents are deeply integrated into the app's core functionality rather than being added as a side-chat.

Highlights

• Unified Actions: Define a function once and deploy it to multiple interfaces (UI, agent, CLI, etc.).
• Comprehensive Runtime: Includes integrated memory, handoffs, and observability.
• Flexible Deployment: Supports various SQL databases via Drizzle and Nitro-compatible hosting.
• Template Gallery: Provides pre-built app templates for analytics, design prototyping, presentations, and content management.