> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/n4ze3m/page-assist/llms.txt
> Use this file to discover all available pages before exploring further.

# Knowledge Base

> Upload documents and chat with your own data using RAG technology

The Knowledge Base feature allows you to upload your own documents and have intelligent conversations about their content. Using RAG (Retrieval Augmented Generation), Page Assist creates searchable embeddings of your documents and retrieves relevant information to answer your questions.

<Warning>
  Knowledge bases are stored entirely in your browser's local storage. Large files may impact browser performance. Consider file size and quantity when building your knowledge base.
</Warning>

## Supported File Types

Page Assist supports the following document formats:

| Format         | Extension | Use Case                                 |
| -------------- | --------- | ---------------------------------------- |
| PDF            | `.pdf`    | Research papers, books, reports, manuals |
| Word Documents | `.docx`   | Essays, documentation, meeting notes     |
| Plain Text     | `.txt`    | Code files, logs, simple notes           |
| CSV            | `.csv`    | Data tables, spreadsheets, lists         |
| Markdown       | `.md`     | Technical docs, notes, README files      |

<Info>
  File processing time depends on document size and complexity. PDFs with images or complex formatting may take longer to process.
</Info>

## Prerequisites

Before using the Knowledge Base feature, you need to configure an embedding model:

<Steps>
  <Step title="Choose an Embedding Model">
    Embedding models convert text into vector representations for semantic search.

    **Recommended Models**:

    * `nomic-embed-text` (Ollama) - Best all-around choice
    * `mxbai-embed-large` (Ollama) - High quality
    * `all-minilm` (Ollama) - Fast and lightweight
    * `text-embedding-3-small` (OpenAI) - Cloud-based option
  </Step>

  <Step title="Install Embedding Model (Ollama)">
    If using local Ollama:

    ```bash theme={null}
    ollama pull nomic-embed-text
    ```
  </Step>

  <Step title="Configure in Settings">
    1. Go to Settings → RAG Settings
    2. Select your embedding model from the dropdown
    3. Configure chunk size (default: 1000)
    4. Configure overlap (default: 200)
    5. Set number of retrieved documents (default: 4)
    6. Click "Save Settings"
  </Step>
</Steps>

<Warning>
  **Important**: Use models specifically designed for text embedding, NOT chat models. Using chat models for embeddings will not work properly.
</Warning>

## Creating a Knowledge Base

Upload and process your documents:

<Steps>
  <Step title="Access Knowledge Management">
    1. Open Page Assist Web UI
    2. Go to Settings
    3. Navigate to "Manage Knowledge"
  </Step>

  <Step title="Add New Knowledge">
    1. Click "Add New Knowledge" button
    2. Enter a descriptive title for your knowledge base
    3. Optionally add custom system and follow-up prompts
    4. Click "Create"
  </Step>

  <Step title="Upload Files">
    1. Click "Upload Files" or drag and drop
    2. Select one or more supported files
    3. Wait for processing - you'll see progress indicators
    4. Files are chunked and embedded automatically
  </Step>

  <Step title="Verify Processing">
    * Green checkmark indicates successful processing
    * View number of chunks created per file
    * Check total embedding count
  </Step>
</Steps>

## Using Knowledge Bases

Once created, use your knowledge bases in conversations:

### In the Web UI

1. Start a new chat or open existing conversation
2. Look for the database/block icon in the input area
3. Click the icon to open knowledge base selector
4. Select one or more knowledge bases to use
5. Selected knowledge bases highlight with a checkmark
6. Type your question and send

### In the Sidebar

1. Open the sidebar (`Ctrl+Shift+Y`)
2. Click the knowledge base icon in the input area
3. Select your knowledge base
4. Ask questions about your documents

<Info>
  You can select multiple knowledge bases simultaneously. The system will search across all selected bases.
</Info>

## Knowledge Base Settings

### RAG Configuration

Optimize how documents are processed and retrieved:

<Accordion title="Chunk Size">
  **Default**: 1000 characters

  Controls how documents are split:

  * **Smaller (500-800)**: Better for precise answers, more chunks
  * **Larger (1200-1500)**: More context per chunk, fewer chunks

  Adjust in Settings → RAG Settings → Chunk Size
</Accordion>

<Accordion title="Chunk Overlap">
  **Default**: 200 characters

  Overlap between consecutive chunks:

  * Prevents information loss at boundaries
  * Higher overlap = more redundancy but better recall
  * Typical range: 100-300 characters

  Adjust in Settings → RAG Settings → Chunk Overlap
</Accordion>

<Accordion title="Retrieved Documents">
  **Default**: 4 documents

  Number of relevant chunks to retrieve:

  * **Fewer (2-3)**: Faster, less context
  * **More (6-8)**: Better coverage, more context

  Adjust in Settings → RAG Settings → Number of Retrieved Documents
</Accordion>

<Accordion title="Splitting Strategy">
  **Default**: RecursiveCharacterTextSplitter

  How text is divided into chunks:

  * **RecursiveCharacterTextSplitter**: Smart splitting by paragraphs, sentences
  * **CharacterTextSplitter**: Simple character-based splitting with custom separator

  Adjust in Settings → RAG Settings → Splitting Strategy
</Accordion>

<Accordion title="Files Per Knowledge Base">
  **Default**: 10 files

  Maximum number of files per knowledge base:

  * Helps manage browser storage
  * Adjust based on file sizes
  * Larger limit may impact performance

  Adjust in Settings → RAG Settings → Total File Per KB
</Accordion>

### Custom Prompts

Create knowledge base-specific prompts:

1. When creating/editing a knowledge base
2. Add custom **System Prompt** to set context
3. Add custom **Follow-up Prompt** for question processing
4. These override global RAG prompts for this knowledge base

**Example System Prompt**:

```text theme={null}
You are an expert assistant analyzing technical documentation. 
Provide accurate, detailed answers citing specific sections.
Context: {context}
```

**Example Follow-up Prompt**:

```text theme={null}
Based on the documentation, {question}
Provide specific examples and page references.
```

## Managing Knowledge Bases

### View and Edit

1. Go to Settings → Manage Knowledge
2. Click on any knowledge base to view details
3. See all uploaded files and their status
4. View total chunks and embeddings
5. Edit title and prompts

### Add More Files

1. Open an existing knowledge base
2. Click "Upload Files"
3. Select additional files
4. New files are processed and added to existing embeddings

### Remove Files

1. Open knowledge base
2. Find the file to remove
3. Click the delete/remove icon
4. Confirm deletion
5. Associated embeddings are automatically removed

### Delete Knowledge Base

<Warning>
  Deleting a knowledge base permanently removes all files and embeddings. This cannot be undone.
</Warning>

1. Go to Settings → Manage Knowledge
2. Find the knowledge base to delete
3. Click the delete icon
4. Confirm deletion
5. All data is removed from browser storage

## Advanced Usage

### Combining Features

<Tabs>
  <Tab title="Knowledge Base + Internet Search">
    Get the best of both worlds:

    1. Enable knowledge base
    2. Enable internet search (globe icon)
    3. Ask questions that reference both your docs and current info
    4. Example: "How does my research compare to recent developments?"
  </Tab>

  <Tab title="Knowledge Base + Webpage">
    Cross-reference documents with web content:

    1. Enable knowledge base
    2. Enable chat with webpage
    3. Compare webpage content with your documents
    4. Example: "How does this article relate to my notes on the topic?"
  </Tab>

  <Tab title="Multiple Knowledge Bases">
    Search across multiple document sets:

    1. Create separate knowledge bases by topic/project
    2. Select multiple bases simultaneously
    3. Ask questions that span multiple domains
    4. Example: "Combine insights from both my research papers and meeting notes"
  </Tab>
</Tabs>

### Optimization Tips

<Tip>
  **Organization**: Create separate knowledge bases for different topics or projects rather than one massive base.
</Tip>

<Tip>
  **File Preparation**: For best results, use well-formatted documents with clear structure and headings.
</Tip>

<Tip>
  **Embedding Model**: Use `nomic-embed-text` for the best balance of speed and quality with local processing.
</Tip>

<Tip>
  **Chunk Size**: For technical docs, use smaller chunks (800). For books/articles, use larger chunks (1200).
</Tip>

## Use Cases

<CardGroup cols={2}>
  <Card title="Research" icon="flask">
    * Analyze research papers
    * Cross-reference studies
    * Extract citations
    * Summarize findings
  </Card>

  <Card title="Documentation" icon="book">
    * Query technical docs
    * Find API references
    * Understand codebases
    * Generate documentation
  </Card>

  <Card title="Learning" icon="graduation-cap">
    * Study textbooks
    * Review course materials
    * Create study guides
    * Generate practice questions
  </Card>

  <Card title="Business" icon="briefcase">
    * Analyze reports
    * Review contracts
    * Extract insights from meeting notes
    * Summarize proposals
  </Card>
</CardGroup>

## Troubleshooting

<Accordion title="File upload fails">
  **Possible causes**:

  * File too large
  * Unsupported format
  * Browser storage full

  **Solutions**:

  * Check file size (keep under 10MB)
  * Verify file extension is supported
  * Clear browser storage or delete old knowledge bases
  * Try splitting large files
</Accordion>

<Accordion title="Processing takes too long">
  **Possible causes**:

  * Large file size
  * Complex PDF formatting
  * Slow embedding generation

  **Solutions**:

  * Use local Ollama for faster embedding generation
  * Reduce chunk size to create fewer embeddings
  * Process files in smaller batches
  * Simplify PDF formatting if possible
</Accordion>

<Accordion title="No embedding model available">
  **Solutions**:

  1. Go to Settings → RAG Settings
  2. Ensure embedding model is selected
  3. For Ollama: Run `ollama pull nomic-embed-text`
  4. For OpenAI: Add API key in OpenAI settings
  5. Refresh and try again
</Accordion>

<Accordion title="Irrelevant results">
  **Possible causes**:

  * Poor chunk retrieval
  * Wrong embedding model
  * Insufficient retrieved documents

  **Solutions**:

  * Increase number of retrieved documents
  * Adjust chunk size and overlap
  * Use better embedding model (nomic-embed-text)
  * Rephrase your question more specifically
  * Try reducing chunk size for more precise matching
</Accordion>

<Accordion title="Storage quota exceeded">
  **Causes**:

  * Too many files
  * Large embeddings
  * Browser storage limit reached

  **Solutions**:

  * Delete unused knowledge bases
  * Remove individual large files
  * Reduce files per knowledge base
  * Clear browser cache
  * Use smaller chunk sizes
</Accordion>

## Performance Considerations

<Warning>
  **Browser Storage**: All embeddings are stored in browser local storage. Each document creates hundreds or thousands of embedding vectors.
</Warning>

### Storage Impact

* **Small doc (10 pages)**: \~1-2 MB of embeddings
* **Medium doc (50 pages)**: \~5-10 MB of embeddings
* **Large doc (200 pages)**: \~20-40 MB of embeddings

### Best Practices

1. **Limit file sizes**: Keep individual files under 10MB
2. **Organize by topic**: Create multiple small knowledge bases vs one large one
3. **Regular cleanup**: Delete unused knowledge bases
4. **Monitor storage**: Check browser storage usage regularly
5. **Use appropriate chunk sizes**: Smaller chunks = more embeddings = more storage

## Privacy and Security

<Info>
  **Local Processing**: All file processing and embedding generation happens locally or via your configured provider. Page Assist has no servers and never stores your data.
</Info>

* Documents are processed in your browser
* Embeddings stored in browser local storage
* Text is only sent to your configured AI provider
* No third-party storage or processing
* Delete knowledge base = permanent removal from your browser

<Note>
  When using cloud embedding providers (OpenAI), your document chunks are sent to their servers for embedding generation.
</Note>

## Next Steps

<CardGroup cols={2}>
  <Card title="Configuration Settings" icon="sliders" href="/configuration/settings">
    Configure embedding models and chunk settings
  </Card>

  <Card title="Prompts" icon="wand-magic-sparkles" href="/features/prompts">
    Create custom prompts for knowledge bases
  </Card>

  <Card title="Internet Search" icon="magnifying-glass" href="/features/internet-search">
    Combine knowledge base with web search
  </Card>

  <Card title="Chat with Webpage" icon="globe" href="/features/chat-with-webpage">
    Cross-reference documents with web content
  </Card>
</CardGroup>