The Knowledge Base Settings tab allows you to configure how your documents are processed and chunked. These settings affect how content is chunked, embedded, and retrieved by your agent.
Chunking Strategy
Choose how your documents will be divided into chunks for processing.
| Strategy | Description |
|---|
| Sentence Chunking | Uses a sentence tokenizer for natural language boundaries. Best for general-purpose content where preserving sentence context is important. |
| Semantic Chunking | Automatically determines chunk boundaries based on semantic meaning. Best for complex or technical documents where semantic context is more important than sentence structure. |
Chunk Configuration
Configure chunk size and overlap for sentence-based chunking. This section applies when Sentence Chunking is selected.
| Setting | Default | Recommendation | Description |
|---|
| Chunk Size | 64 | 64–512 tokens | The maximum size of each chunk. Larger chunks preserve more context but may reduce retrieval precision. Smaller chunks are more precise but may lose context. |
| Chunk Overlap | 6 | 10–20% of chunk size | The number of overlapping tokens between adjacent chunks. Helps maintain context across chunk boundaries. Must be non-negative and less than chunk size. |
Chunk Enhancement Settings
Configure chunk enrichment and LLM model selection for better content processing.
| Setting | Description |
|---|
| Chunk Enrichment | When enabled, uses AI to enhance chunks with additional context for better retrieval accuracy. |
| LLM Model | Select the LLM model to use for chunk enrichment. Required when Chunk Enrichment is enabled. |
If no Chunk Enrichment models are configured in Super Admin, the default model will be used automatically.
Chunk enrichment improves search quality but increases processing time and computational cost. Enable it for complex content where simple chunking isn’t sufficient.
Danger Zone
The actions in this section are irreversible. Proceed with caution.
Saving and Applying Settings
| Action | Description |
|---|
| Save Settings | Saves your configuration. Applies only to new documents added or synced going forward. |
| Reset | Resets the form to the last saved configuration. |
| Apply to All Documents | Re-indexes all existing documents in your knowledge base with the current settings. Use this after saving to apply changes to previously uploaded content. |
Applying settings to all documents can take significant time depending on the number of documents in your knowledge base. This process cannot be cancelled once started.
Best Practices
Chunking Strategy
- Start with Sentence Chunking for most use cases.
- Switch to Semantic Chunking for complex or technical documents where sentence boundaries don’t align with meaningful content divisions.
Chunk Size
- Start with the default (64) as a baseline.
- Increase for content that requires more context per chunk.
- Decrease when precise, targeted retrieval is more important than context.
Chunk Overlap
- Keep overlap at 10–20% of your chunk size.
- Increase overlap when critical information may span chunk boundaries.
Chunk Enrichment
- Enable for complex content where retrieval quality needs improvement.
- Choose a model with strong reasoning capabilities for best results.
- Monitor processing time after enabling — enrichment adds computational cost.
