agent-native

Agent-Native is an open-source TypeScript framework for building agentic applications that combine autonomous agent work with a purpose-built user interface. Its central idea is that each capability is defined once as an "action": the agent invokes it as a tool, while the UI calls the same function from code, so both paths share validation, permissions and implementation. Key concepts described in the README: - Shared actions: one definition powers the UI, agent, HTTP, MCP, A2A and CLI surfaces. The example shows a `defineAction` call with a Zod schema, an HTTP method and a `run` function; the agent receives it as a tool and React can call it via `useActionQuery`. - Shared data: work performed by the agent appears in the UI, and work done in the UI is available to the agent. - Shared application state: the agent receives relevant UI context such as the current page, selected record or active view. The agent does not click through the UI; it operates through the same action layer. Included capabilities listed in the README: - Agent chat for delegating work, asking questions and reviewing results in the same UI. - Authentication and permissions controlling access to shared work. - Skills and memory for reusable expertise and persistent context. - Automations that run agent work on schedules or events. - Agent teams that delegate to specialist agents in the same workspace or across connected agents. - A PostgreSQL backend, with PostgreSQL in production and PGlite for local development on Nitro-compatible hosts. The project states that users can bring their own LLM, SQL database, tools and infrastructure, and that everything built stays theirs. Quick start is via `npx --yes @agent-native/core@latest create my-agent --standalone --template chat`, with a getting started guide linked from the documentation site. The README also presents a gallery of open-source example agents that can be used as starting points or references, including Clips (recording and understanding meetings, screens and voice notes), Design (generating and refining interactive designs), Slides (creating and editing on-brand presentations), Analytics (asking questions of data and building dashboards), Calendar (finding time, scheduling events and managing bookings), Mail (prioritizing email, drafting replies and following up), Assets (creating and organizing on-brand media), Content (drafting, organizing and publishing content) and Plans (creating and reviewing visual plans with diagrams, wireframes and prototypes). Documentation and community links point to the project documentation site and a Discord server. Contributing guidance refers to a DEVELOPMENT.md file for local setup, workspace structure and guard scripts. The project is released under the MIT license.