Agent Native Analytics

What Agent-Native Analytics does

Agent-Native Analytics brings analytics work into an agent-driven application. It connects data sources including Amplitude and Mixpanel, allowing users to ask questions about their data and create charts or dashboards through prompts. The analytics experience is presented as one of the open-source applications built with the Agent-Native framework.

The broader framework is designed around capabilities called actions. An action can be used by an agent as a tool and by a user interface from application code. This shared action layer keeps the implementation, validation, and permissions consistent across interaction surfaces.

How it works

Agent-Native applications share three parts of the working context. Actions are available to both the agent and the UI; data produced through one surface can appear in the other; and relevant application state, such as the current page, selected record, or active view, can be supplied to the agent. Agents operate through the action layer rather than clicking through the interface.

A defined action can be exposed through several interfaces at once: the application UI, HTTP, MCP, A2A, and the command line. In practice, this lets an MCP client invoke the same capability that the application uses internally. The repository describes actions as TypeScript modules with a description, a Zod input schema, optional HTTP settings, and an asynchronous implementation.

Setup and configuration

The project provides a command-line quick start for creating a standalone Agent-Native application:

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

The resulting application can be developed as an agent with a dedicated UI. The framework supports PostgreSQL for production deployments and PGlite for local development on Nitro-compatible hosts. Authentication and permissions are available for controlling access to shared work.

Agent-Native is built as an open-source TypeScript framework, so using the analytics experience involves working within that application and action model rather than configuring an isolated analytics-only endpoint.

Tools and capabilities

• Ask questions about connected analytics data.
• Prompt for charts and build dashboards.
• Expose shared actions as MCP tools.
• Reach the same actions through HTTP, A2A, and the CLI.
• Share application state and action results between the agent and UI.
• Add scheduled or event-driven automations.
• Provide reusable skills and persistent agent context.
• Delegate work to specialist agents through agent teams.

The repository also includes related Agent-Native applications for meeting and voice-note analysis, interactive design, presentations, calendars, and email. These examples demonstrate the framework’s broader application model; the analytics use case focuses on questions, charts, and dashboards.