> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/poweroutlet2/openground/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> On-device RAG system for AI-powered documentation search

# Welcome to OpenGround

OpenGround lets you give controlled access to documentation to AI agents. Everything happens on-device.

<Info>
  OpenGround is an on-device RAG system that extracts documentation from git repos and sitemaps, embeds it for semantic search, and exposes it to AI agents via MCP.
</Info>

## What is OpenGround?

OpenGround bridges the gap between your AI coding assistant and official documentation. Instead of relying on potentially outdated training data, your AI agents can search accurate, up-to-date documentation stored locally on your machine.

<CardGroup cols={2}>
  <Card title="On-Device Processing" icon="laptop" iconType="duotone">
    All data stays on your machine. No cloud services, no API keys, complete privacy.
  </Card>

  <Card title="Hybrid Search" icon="magnifying-glass" iconType="duotone">
    Combines semantic vector search with BM25 full-text search for accurate results.
  </Card>

  <Card title="MCP Integration" icon="plug" iconType="duotone">
    Seamlessly integrates with Claude Code, Cursor, OpenCode, and other MCP-compatible agents.
  </Card>

  <Card title="Multi-Version Support" icon="code-branch" iconType="duotone">
    Store and query multiple versions of the same library simultaneously.
  </Card>
</CardGroup>

## Architecture

```
      ┌─────────────────────────────────────────────────────────────────────┐
      │                           OPENGROUND                                │
      ├─────────────────────────────────────────────────────────────────────┤
      │                                                                     │
      │       SOURCE                  PROCESS              STORAGE/CLIENT   │
      │                                                                     │
      │    ┌──────────┐      ┌───────────┐   ┌──────────┐   ┌──────────┐    │
      │    │ git repo ├─────>│  Extract  ├──>│  Chunk   ├──>│ LanceDB  │    │
      │    |   -or-   |      │ (raw_data)│   │   Text   │   │ (vector  │    │
      │    │ sitemap  │      └───────────┘   └──────────┘   │  +BM25)  │    │
      │    │   -or-   │                           │         └────┬─────┘    │
      │    │ local dir│                           │              │          │
      │    └──────────┘                           │              │          │
      │                                           ▼              │          │
      │                                    ┌───────────┐         │          │
      │                                    │   Local   |<────────┘          │
      │                                    │ Embedding │         │          │
      │                                    │   Model   │         ▼          │
      │                                    └───────────┘  ┌─────────────┐   │
      │                                                   │ CLI / MCP   │   │
      │                                                   │  (hybrid    │   │
      |                                                   |   search)   |   |
      │                                                   └─────────────┘   │
      │                                                                     │
      └─────────────────────────────────────────────────────────────────────┘
```

OpenGround follows a simple pipeline:

1. **Source** - Extract docs from git repos, sitemaps, or local directories
2. **Process** - Chunk text and generate embeddings using local models
3. **Storage** - Store in LanceDB with vector + BM25 indices
4. **Query** - Search via CLI or expose to AI agents through MCP

## Key Features

### Privacy First

All processing happens on your machine. No data leaves your device. No cloud dependencies.

### Flexible Sources

Extract documentation from:

* **Git repositories** - Clone specific versions, sparse checkout for docs folders
* **Sitemaps** - Crawl documentation sites automatically
* **Local directories** - Index your own documentation or notes

### Powerful Search

Hybrid search combines:

* **Semantic search** - Understands meaning and context
* **BM25** - Traditional keyword matching for precise queries

### AI Agent Integration

Seamless integration with:

* Claude Code
* Cursor
* OpenCode
* Any MCP-compatible agent

### Version Management

* Store multiple versions of the same library
* Query specific versions
* Incremental updates with content hash detection

### Multiple Embedding Backends

* **Sentence-Transformers** - Full GPU/MPS/CPU support, larger package
* **FastEmbed** - Lightweight, CPU-optimized
* **FastEmbed-GPU** - Experimental CUDA support

## Use Cases

<Accordion title="AI Coding Assistants">
  Give your AI assistant access to official documentation without polluting your conversation context. Ask questions like "How do I use React's useEffect hook?" and get answers directly from the React documentation.
</Accordion>

<Accordion title="Multi-Project Development">
  Work with multiple versions of the same framework? Store React 17 and React 18 docs side-by-side, query the version you're currently using.
</Accordion>

<Accordion title="Offline Development">
  Download documentation once, use it forever offline. Perfect for planes, remote locations, or when you want to avoid distractions.
</Accordion>

<Accordion title="Private Documentation">
  Index your company's internal documentation and make it searchable for AI agents, without sending it to the cloud.
</Accordion>

## Supported File Formats

When extracting from git repos or local directories, OpenGround supports:

* Markdown (`.md`, `.mdx`)
* reStructuredText (`.rst`)
* Plain text (`.txt`)
* Jupyter Notebooks (`.ipynb`)
* HTML (`.html`, `.htm`)

## How It Works

OpenGround uses a **Retrieval-Augmented Generation (RAG)** approach:

1. **Extract** - Download documentation from various sources
2. **Chunk** - Split documents into manageable pieces (800 chars with 200 char overlap)
3. **Embed** - Generate vector embeddings using BGE-small-en-v1.5 model
4. **Index** - Store in LanceDB with both vector and BM25 full-text indices
5. **Query** - Search using hybrid approach that combines semantic and keyword matching
6. **Serve** - Expose results via CLI or MCP server for AI agents

## Next Steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Get up and running in 5 minutes
  </Card>

  <Card title="Installation" icon="download" href="/installation">
    Detailed installation instructions
  </Card>

  <Card title="Add Documentation" icon="book" href="/guides/add-from-git">
    Learn how to add your first library
  </Card>

  <Card title="MCP Integration" icon="plug" href="/integration/mcp-overview">
    Connect to your AI coding assistant
  </Card>
</CardGroup>