1. Enable AI
1. Enable AI
- Locate the “Use AI” toggle in the settings.
- Switch it to the enabled state (purple).
2. Select the Model Provider
2. Select the Model Provider
- Choose the AI model provider from the “Model Provider” dropdown menu:
- If using OpenAI, ensure “OpenAI” is selected.
- If using Ollama, ensure “Ollama” is selected.
3. Select the AI Model
3. Select the AI Model
- Choose a suitable model for your task from the “AI Model” dropdown.
4. Set the Embedding Model
4. Set the Embedding Model
- Select the embedding model from the “Embedding Model” dropdown.
5. Input the API Key
5. Input the API Key
- Enter your OpenAI API key in the “API Key” input field:
- Ensure that your key is correct and has the necessary permissions.
6. Set the API Endpoint
6. Set the API Endpoint
- Configure the API endpoint in the “API Endpoint” input field:
- Ensure the address is valid and matches your API configuration.
7. Rebuild the Embedding Index
7. Rebuild the Embedding Index
- If you need to rebuild the embedding index:
- Click the “Rebuild” button to regenerate the embeddings.
- If a forced rebuild is required, click “Force Rebuild.”
Ensure the correctness of the API key and endpoint to avoid functionality issues.
Advanced
Parameters Explained
Top K
Top K
- Determines the number of most relevant chunks to consider
- Range: 1-20 (recommended: 2-5)
- Higher values include more context but may reduce relevance
Score
Score
- Minimum similarity score threshold for context inclusion
- Range: 0.0-2.0 (recommended: 0.7-0.9)
- Higher values ensure more relevant matches
- Lower values include more potential matches
Embedding Lambda
Embedding Lambda
- Controls the balance between semantic and lexical similarity
- Range: 0.0-1.0 (recommended: 0.2-0.4)
- Higher values favor semantic meaning
- Lower values favor exact word matches
Best Practices
For optimal results:
- Start with default values (Top K: 2, Score: 0.8, Lambda: 0.3)
- Adjust gradually based on response quality
- Monitor response relevance and accuracy
Extreme parameter values may lead to:
- Very limited or irrelevant responses
- Increased processing time
- Reduced accuracy