Agent-Native is an open-source framework designed for building applications where AI agents and user interfaces operate with shared actions, state, and context. Written in TypeScript, the framework enables developers to define business logic once and deploy it across multiple surfaces including UI components, agent interactions, HTTP endpoints, MCP protocols, agent-to-agent communication, and command-line interfaces.
The framework's core architecture centers on three main pillars. Actions serve as the fundamental unit of work that can be invoked from any application surface without duplication. The agent runtime bundles together chat capabilities, tool definitions, skill management, memory systems, job scheduling, observability features, and handoff mechanisms into a cohesive runtime environment. Backend flexibility is achieved through support for any Drizzle-compatible SQL database and any Nitro-compatible hosting platform, allowing developers to choose their preferred infrastructure.
Agent-Native ships with a collection of template applications that demonstrate the framework's capabilities across different domains. The Clips application combines screen recording with automatic transcription and browser debug log capture, enabling agents to analyze recorded sessions and identify bugs. The Plans template provides visual planning and recap functionality for coding agents, allowing them to outline approaches before implementation and review changes with diagrams and annotations afterward. The Design template facilitates agent-driven interactive HTML prototype generation with variant comparison and control refinement. Additional templates include Content for MDX editing with agent-assisted writing, Slides for React-based presentation generation, Analytics for connecting data sources and building dashboards, and Mail, Calendar, and Forms applications that can be combined in a shared workspace with unified authentication.
The quick start process accommodates different development preferences through a single command that guides users toward their chosen starting point. Developers can clone complete applications into a workspace, start with a minimal chat interface, or begin with a headless action-first application. Command-line flags provide shortcuts for experienced users who prefer to skip interactive prompts.
The framework is distributed under the MIT license and maintains comprehensive documentation at agent-native.com, with community support available through a Discord server for developers building with the platform.
