Developing OmniMind

This document contains instructions for building, running, and modifying the OmniMind plugin from source. If you are an AI agent assisting with the codebase, please refer to the AGENTS.md file for architectural rules and constraints.

🛠️ Tech Stack

• Backend: Node.js, TypeScript, @lmstudio/sdk
• Frontend: React, Vite, CSS (No-Tailwind)
• Vector Database: LanceDB
• PDF Processing: mupdf
• Orchestration: Native LM Studio SDK (.act() loop)

🏗️ Architecture Deep Dive

• LanceDB Vector Store: Uses LanceDB for blazing-fast, serverless semantic search. We leverage LanceDB's optimistic concurrency controls to ensure the ingestion queue and search API don't collide.
• Premium Web Control Panel: The UI is built using Vite + React. It allows developers and power-users to browse the raw LanceDB vector chunks directly to debug ingestion logic.
• Native SDK Orchestration: We rely exclusively on LM Studio's built-in agentic .act() workflows. The agent inherently decides when to query the knowledge graph or answer directly, eliminating the need for heavy, error-prone third-party orchestration frameworks like LangChain.

🚀 Setting Up the Developer Environment

If you want to modify the plugin from source:

• Clone the repository:

• Install dependencies:

• Build the project:

🏃 Running in Development Mode

To start the plugin server with live-reloading for local development:

This will automatically compile changes and connect the plugin to your running LM Studio instance.

🧪 Running Tests

OmniMind uses the native node:test runner. To execute the test suite: