Lumina Note: Markdown App with AI Agent Integration

What changed

Lumina Note has been released as a modern Markdown note-taking application designed to keep notes human-owned and editable while embedding an AI agent capable of operating across the entire note vault. The application's core philosophy is to maintain plain Markdown files on disk, ensuring users can access, edit, and back up their data without relying on Lumina or specific model providers. The AI agent, powered by the opencode SDK, can perform tasks such as drafting pages, reorganizing folders, summarizing PDFs, connecting related notes, generating flashcards, and cleaning up knowledge bases, all with user permission.

Key features include a live preview for Markdown editing, reading mode, WikiLinks for creating bidirectional connections between notes, backlinks, and graph views to visualize note relationships. Lumina Note offers flexible AI routing, allowing users to connect to various model providers. These include OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Moonshot Kimi, Zhipu GLM, MiMo, Groq, OpenRouter, and local models via Ollama. Users can also connect to OpenAI-compatible endpoints or opt for a licensed Lumina Cloud service. The application is built as an Electron desktop app using React and TypeScript, with its Markdown editor based on CodeMirror. The agent layer utilizes an embedded opencode server and SDK, with Lumina handling app settings, vault access, and permission flows.

For developers, Lumina Note provides a plugin system in developer preview, enabling extensions to the app's functionality. Building from source requires Node.js 22+, npm, and Bun for bundling the opencode server. The setup involves cloning the repository, installing dependencies, and setting up the opencode server bundle before running the development server.

Why it matters for builders

Lumina Note offers a compelling foundation for developers looking to integrate AI capabilities into note-taking and knowledge management workflows. The application's human-first approach to data ownership, combined with its agent-native assistance, provides a unique environment for building AI-powered tools. The flexible AI routing allows developers to experiment with different LLM providers and local models, adapting to various project needs and cost considerations. Furthermore, the extensible plugin system and the ability to build on the opencode SDK empower developers to create custom agent skills and commands, tailoring the application to specific use cases.

Practical impact

Builders can install Lumina Note on Windows, macOS, and Linux. The quick start guide outlines the process of setting up a vault, configuring an AI model provider, and beginning to use the AI agent for tasks. Developers interested in extending Lumina Note can explore the plugin ecosystem documentation and the agent runtime setup. The project is licensed under the Apache License 2.0, encouraging contributions and modifications. The availability of builds from Releases and the option to build from source provide immediate access for experimentation and development.

Caveats and source limits

The source material indicates a developer-preview status for the plugin system, suggesting potential changes and a need for caution when building extensions. While Lumina Note supports a wide array of AI model providers, specific performance benchmarks or cost comparisons for these integrations are not provided. The documentation for mobile companions is referenced but not detailed within the provided excerpts. The exact capabilities and limitations of the agent when operating across a user's entire vault are not exhaustively detailed, beyond the examples given.