agent-native: Build Agentic Apps with Shared UI and Actions

Overview

Agent-Native is a TypeScript framework for building agentic applications with a purpose-built user interface. Its central design is to define capabilities as shared actions: agents can call them as tools, while application code can invoke the same implementations.

This approach is useful when an agent needs to work with the same records and application context as a person, without automating clicks through the interface. The framework also provides surfaces and infrastructure for chat, authentication, memory, automation, agent teams, and persistent data.

Key Features

• Define actions with schemas and reuse them from the agent and application UI.
• Expose actions through HTTP, MCP, A2A, and the CLI.
• Share data and relevant application state between agents and the UI.
• Build agent chat experiences for delegating work and reviewing results.
• Add authentication and permissions to shared work.
• Provide agent skills, persistent memory, scheduled or event-based automations, and agent teams.
• Use PostgreSQL in production or PGlite for local development.
• Start from open-source example agents for tasks such as analytics, content, scheduling, and design.

Use Cases

• Product teams building an AI feature that must read or change the same application data as its users.
• Developers creating internal tools where people need to inspect, edit, or approve agent work in a familiar UI.
• Teams exposing capabilities across a web interface, agent tools, and APIs without duplicating action logic.
• Builders prototyping specialized agents for workflows such as email, analytics, presentations, or content management.

Project Facts

• Language: TypeScript
• Stars: 7k
• Forks: 637
• Topics: agent-native, agents, ai, react, typescript
• Archived: No

Getting Started

Create a standalone chat template with:

npx --yes @agent-native/core@latest create my-agent --standalone --template chat

See the README and getting started guide for setup and framework details.

Considerations

• The repository metadata does not provide a license value, although the README states that the project is MIT licensed.
• The framework combines an agent with an application UI, so it is a better fit for products needing shared actions and state than for a standalone chat-only bot.
• Production use involves choosing and configuring an LLM, database, and infrastructure. The README describes PostgreSQL for production and PGlite for local development, but does not specify a single required model provider.