Ashok Gelal
3.5 KiB
title, description, navTruncate
| title | description | navTruncate |
|---|---|---|
| Advanced Knowledge Stack Settings | Fine-tune document processing and search quality | false |
Learn how to optimize your Knowledge Stack's search accuracy and processing with these advanced settings.
Processing Settings
::image-plain{src="/how-to/create-knowledge-stack/chunk-settings-interface.webp" alt="Chunk settings interface"} ::
Access these settings by clicking the gear icon at the bottom of your Knowledge Stack:
Embedding Model
Choose how Msty understands your content:
- Local: Private, offline processing
- Remote: Enhanced accuracy via OpenAI
- Click to see all configured models
Splitter Type
Choose how to divide your content:
- Recursive Character: Like chapters in a book
- Best for normal documents
- Keeps paragraphs together
- Works well with mixed content
- Sentence: Like individual sentences
- Perfect for technical docs
- Keeps code blocks intact
- Better for precise answers
Chunk Settings
Control how Msty processes your text:
-
Chunk Size
- Low: Quick, specific answers
- Medium: Balanced (recommended)
- High: More context
- Highest: Full context
-
Chunk Overlap
- Low: Minimal connection
- Medium: Good balance
- High: Strong connection
- Highest: Maximum context
-
Ignore Small Chunks
- Default: 50 characters
- Filters out headers, footers
- Skips single-line comments
- Removes short metadata
::alert{type="tip" icon="tabler:bulb"} Start with Medium settings - they work great for most documents! Adjust only if you notice issues with answer quality. ::
Search Settings
::image-plain{src="/how-to/create-knowledge-stack/search-settings-interface.webp" alt="Search settings interface"} ::
Access these by clicking the sliders icon in the top right:
Results Control
-
Number of Chunks
- Default: 15 chunks
- Adjust based on needs
- More chunks = broader context
- Fewer = focused answers
-
Similarity Threshold
- Low to Highest options
- Controls match strictness
- Higher = more precise matches
- Lower = broader results
Jina AI Reranking
Enhance search accuracy:
- Free API key from jina.ai/reranker
- Improves result ranking
- Better handles complex queries
::alert{type="warning" icon="tabler:alert-triangle"} Privacy Note: Jina AI processes search queries and metadata to improve results. Check their privacy policy for details. ::
Managing Changes
::image-plain{src="/how-to/create-knowledge-stack/compose-save-as-draft-buttons.webp" alt="Compose and save buttons"} ::
Save Options
- Save Draft: Store unfinished work
- Compose: Activate your stack
- Three-dot Menu:
- "Compose new changes": Update content
- "Recompose Stack": Apply setting changes
When to Adjust Settings
-
Content Issues
- Long documents → Increase chunk size
- Missing context → Increase overlap
- Too much noise → Raise similarity threshold
- Short answers → Add more chunks
-
Performance Tips
- Test with real questions
- Change one setting at a time
- Note what works best
- Save successful configurations
::alert{type="info" icon="tabler:info-circle"} Remember: You can adjust search settings per chat session in the Knowledge Stack selector! ::
::alert{type="success" icon="tabler:hand-love-you"} Thanks to our wonderful Discord user @Tdwag fo contributing almost all the content for this guide! ::