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

# RAG Search

> Search documents with AI locally in Vault workspaces using Retrieval-Augmented Generation

## Overview

RAG (Retrieval-Augmented Generation) Search provides AI-powered document search that runs completely offline in Vault workspaces. Unlike cloud-based AI features, RAG Search processes your documents locally, ensuring privacy and security while delivering intelligent search results with AI-generated overviews.

<Info>
  RAG Search is exclusive to Vault workspaces, introduced in v0.9.5. It combines local AI with vector embeddings for private, offline document intelligence.
</Info>

## What Is RAG?

Retrieval-Augmented Generation (RAG) combines two AI capabilities:

1. **Retrieval** - Finding relevant documents using semantic search
2. **Generation** - Creating AI summaries based on retrieved content

### How It Works

<Steps>
  <Step title="Document Indexing">
    Your documents are processed locally and converted into vector embeddings (numerical representations of meaning).
  </Step>

  <Step title="Semantic Search">
    When you search, the query is converted to a vector and compared against document vectors to find semantically similar content.
  </Step>

  <Step title="AI Generation">
    Retrieved documents are sent to your local AI model (via Ollama) to generate a contextual summary with citations.
  </Step>
</Steps>

<Note>
  All processing happens on your device. No data is sent to the cloud, ensuring complete privacy.
</Note>

## Setting Up RAG Search

### Prerequisites

<Steps>
  <Step title="Create a Vault Workspace">
    RAG Search only works in Vault workspaces. Create a new Vault workspace from Settings → Workspaces.
  </Step>

  <Step title="Install Ollama">
    Download and install [Ollama](https://ollama.ai) for local AI processing.

    ```bash theme={null}
    # macOS
    brew install ollama

    # Linux
    curl -fsSL https://ollama.ai/install.sh | sh
    ```
  </Step>

  <Step title="Download AI Models">
    Pull the required models for embeddings and generation:

    ```bash theme={null}
    # Embedding model for document indexing
    ollama pull nomic-embed-text

    # Generation model for summaries
    ollama pull llama2
    # or
    ollama pull mistral
    ```
  </Step>

  <Step title="Configure Embedding Model">
    In AppFlowy Settings → AI, select your preferred embedding model from the available options.
  </Step>
</Steps>

### Supported File Types

RAG Search can process:

* **PDF files** - Extracts text and indexes content
* **Markdown files** (.md) - Indexes structured content
* **Text files** (.txt) - Indexes plain text
* **AppFlowy pages** - Native document support

<Tip>
  Upload files to your Vault workspace by dragging and dropping them into the file manager or importing them through the menu.
</Tip>

## Using RAG Search

<Steps>
  <Step title="Open Search in Vault">
    Press `Cmd/Ctrl + K` to open the search panel in your Vault workspace.
  </Step>

  <Step title="Ask a Question">
    Type a natural language question or search query:

    * "What are the key findings in the research papers?"
    * "Summarize the meeting notes from this week"
    * "Find information about deployment procedures"
  </Step>

  <Step title="Review AI Overview">
    RAG Search returns:

    * **AI-generated summary** - Comprehensive answer to your question
    * **Key highlights** - Important points extracted from documents
    * **Source citations** - Links to specific files and pages
  </Step>

  <Step title="Explore Sources">
    Click source links to view the original documents and verify information.
  </Step>
</Steps>

## RAG Search Features

### Semantic Search

RAG Search understands meaning, not just keywords:

* **Conceptual matching** - Finds documents about "machine learning" when you search for "AI training"
* **Synonym recognition** - Matches "purchase" with "buy", "acquire", "procurement"
* **Context awareness** - Understands multi-word concepts and relationships

### Chat with Files

Ask questions about uploaded documents:

```plaintext theme={null}
Question: "What are the main conclusions in the research PDF?"

RAG Search:
1. Finds relevant sections in the PDF
2. Extracts key conclusions
3. Generates a summary with page references
4. Provides direct quotes with citations
```

### Multi-Document Synthesis

Combine information from multiple files:

* **Cross-document analysis** - Compare findings across multiple PDFs
* **Timeline building** - Order events from different markdown notes
* **Theme identification** - Find common themes across text files
* **Comprehensive answers** - Synthesize information from multiple sources

<CardGroup cols={2}>
  <Card title="PDF Search" icon="file-pdf">
    Extract and search content from PDF documents with page-level citations
  </Card>

  <Card title="Markdown Search" icon="markdown">
    Search through structured markdown notes with section references
  </Card>

  <Card title="Text Search" icon="file-lines">
    Index and search plain text files with line number citations
  </Card>

  <Card title="Workspace Search" icon="folder">
    Search all AppFlowy pages in your Vault workspace
  </Card>
</CardGroup>

## Embedding Models

Choose the right embedding model for your needs:

### Available Models

| Model                 | Size    | Speed     | Quality | Best For                           |
| --------------------- | ------- | --------- | ------- | ---------------------------------- |
| **nomic-embed-text**  | \~274MB | Fast      | High    | General purpose, multilingual      |
| **all-minilm**        | \~45MB  | Very Fast | Good    | Quick indexing, limited resources  |
| **mxbai-embed-large** | \~669MB | Slower    | Highest | Maximum quality, powerful hardware |

### Switching Embedding Models

<Warning>
  Changing embedding models requires re-indexing all documents, which may take time depending on your content volume.
</Warning>

<Steps>
  <Step title="Go to AI Settings">
    Navigate to Settings → AI → Vault Configuration
  </Step>

  <Step title="Select New Model">
    Choose a different embedding model from the dropdown
  </Step>

  <Step title="Trigger Re-indexing">
    AppFlowy automatically re-indexes all documents with the new model
  </Step>

  <Step title="Wait for Completion">
    Monitor the indexing progress indicator until complete
  </Step>
</Steps>

## RAG Search Best Practices

<CardGroup cols={2}>
  <Card title="Ask Complete Questions" icon="comment">
    Use full sentences like "What were the Q1 revenue figures?" rather than just "revenue"
  </Card>

  <Card title="Be Specific" icon="crosshairs">
    Include context: "What did the design review document say about mobile UX?"
  </Card>

  <Card title="Verify Sources" icon="shield-check">
    Always check source documents for critical information and exact details
  </Card>

  <Card title="Organize Files" icon="folder-tree">
    Use clear file names and organize content in folders for better context
  </Card>
</CardGroup>

### Optimizing Document Indexing

**For better RAG Search results:**

* Use descriptive file names that indicate content
* Structure markdown files with clear headings
* Keep related documents in the same folders
* Add metadata (dates, authors, topics) to file names
* Break large documents into logical sections

### Query Optimization

**Write queries that get better results:**

✅ **Good queries:**

* "Summarize the key features discussed in the product spec"
* "What budget was allocated for marketing in Q1?"
* "Compare the two proposal documents and highlight differences"

❌ **Less effective queries:**

* "features"
* "budget"
* "proposals"

## Privacy and Security

### What Stays Local

<Info>
  RAG Search in Vault workspaces is 100% private. Everything runs on your device:
</Info>

* **Document content** - Never leaves your device
* **Embeddings** - Stored locally in your Vault database
* **AI processing** - Runs through local Ollama instance
* **Search queries** - Processed entirely offline
* **Generated summaries** - Created by your local AI model

### Data Storage

RAG Search stores:

* **Vector embeddings** - In local SQLite database
* **Document metadata** - File names, locations, timestamps
* **No external services** - Zero cloud dependencies for search

<Note>
  Vault workspaces are designed for sensitive data. RAG Search maintains this privacy guarantee by keeping all AI operations local.
</Note>

## Performance Considerations

### Hardware Requirements

RAG Search performance depends on your hardware:

**Minimum:**

* 8GB RAM
* 2GB free disk space
* Multi-core processor

**Recommended:**

* 16GB+ RAM
* 10GB+ free disk space (for larger models)
* Apple Silicon M1/M2 or modern x86 processor with AVX2

### Indexing Speed

Document indexing time varies by:

* **File size** - Larger PDFs take longer
* **File count** - More files = longer initial indexing
* **Model size** - Larger embedding models are slower
* **Hardware** - Faster CPU/GPU speeds up processing

<Tip>
  Index documents in batches. Add files to your Vault gradually rather than importing hundreds at once.
</Tip>

### Search Speed

Once indexed, RAG Search is fast:

* **Query matching** - Near-instant with vector search
* **AI generation** - 2-10 seconds depending on model and context
* **No network latency** - Offline operation is consistently fast

## Troubleshooting

### RAG Search Not Available

**Problem:** RAG Search option doesn't appear in search.

**Solutions:**

1. Verify you're in a Vault workspace (not a regular workspace)
2. Check that Ollama is installed and running: `ollama list`
3. Ensure embedding model is downloaded: `ollama pull nomic-embed-text`
4. Restart AppFlowy after installing Ollama

### Slow Indexing

**Problem:** Document indexing takes too long.

**Solutions:**

1. Use a smaller, faster embedding model (all-minilm)
2. Index fewer documents at once
3. Close other applications to free up resources
4. Check available disk space
5. Consider upgrading hardware for better performance

### Poor Search Results

**Problem:** RAG Search returns irrelevant results.

**Solutions:**

1. Rewrite queries with more specific details
2. Check that documents are properly indexed (Settings → AI)
3. Try a different embedding model for better quality
4. Verify uploaded files contain searchable text (PDFs may have image-only content)
5. Re-index documents if search quality has degraded

### AI Generation Errors

**Problem:** Search finds documents but AI fails to generate summary.

**Solutions:**

1. Verify Ollama is running: `ollama serve`
2. Check generation model is available: `ollama list`
3. Download recommended model: `ollama pull llama2`
4. Try a different generation model
5. Check logs for specific error messages

## RAG Search vs. Cloud AI Search

| Feature        | RAG Search (Vault)     | Cloud AI Search          |
| -------------- | ---------------------- | ------------------------ |
| **Privacy**    | 100% local             | Cloud-processed          |
| **Internet**   | Works offline          | Requires connection      |
| **Speed**      | Fast (local)           | Depends on network       |
| **Models**     | Ollama models          | GPT, Claude, etc.        |
| **File Types** | PDF, MD, TXT           | AppFlowy pages only      |
| **Setup**      | Requires Ollama        | No setup                 |
| **Cost**       | Free (local resources) | May require subscription |

## Advanced RAG Configuration

### Embedding Model Parameters

Adjust embedding behavior in Settings → AI → Advanced:

* **Chunk size** - Size of document sections for indexing (default: 512 tokens)
* **Chunk overlap** - Overlap between chunks for context (default: 128 tokens)
* **Top K results** - Number of document chunks to retrieve (default: 5)
* **Score threshold** - Minimum similarity score for results (default: 0.5)

### Generation Model Selection

Choose the best generation model for your needs:

| Model         | Size  | Speed  | Quality | Use Case           |
| ------------- | ----- | ------ | ------- | ------------------ |
| **llama2**    | 3.8GB | Medium | Good    | General purpose    |
| **mistral**   | 4.1GB | Medium | High    | Better reasoning   |
| **mixtral**   | 26GB  | Slow   | Highest | Maximum quality    |
| **codellama** | 3.8GB | Medium | Good    | Code-focused tasks |

### Custom RAG Workflows

Combine RAG Search with other features:

1. **Research workflow** - Upload PDFs → RAG Search → Save summaries to pages
2. **Meeting notes** - Index markdown notes → Search across meetings → Generate weekly summaries
3. **Documentation** - Index text files → Ask implementation questions → Generate guides
4. **Code analysis** - Index code files → Search for patterns → Get explanations

## Related Features

<CardGroup cols={2}>
  <Card title="Vault Workspaces" icon="vault" href="/advanced/vault-workspace">
    Learn about private, offline Vault workspaces
  </Card>

  <Card title="AI Overviews" icon="lightbulb" href="/ai/ai-overviews">
    Cloud-based AI summaries for regular workspaces
  </Card>

  <Card title="AI Search" icon="magnifying-glass" href="/ai/ai-search">
    Natural language search in cloud workspaces
  </Card>

  <Card title="AI Chat" icon="comments" href="/ai/ai-chat">
    Interactive AI conversations with document sources
  </Card>
</CardGroup>