mirror of
https://github.com/cloudstack-llc/msty-docs.git
synced 2026-08-27 10:11:21 -04:00
Adam Miedema
83 lines
2.6 KiB
Markdown
83 lines
2.6 KiB
Markdown
---
|
|
title: Understanding RAG in Knowledge Stacks
|
|
description: Learn how Retrieval Augmented Generation works in Msty
|
|
navTruncate: false
|
|
---
|
|
|
|
::MstyStudioDocsNotice
|
|
::
|
|
|
|
## What is RAG?
|
|
|
|
RAG (Retrieval Augmented Generation) is the technology that powers Knowledge Stacks in Msty. It's important to understand that RAG doesn't "train" or "teach" the AI new information - instead, it's more like giving the AI a temporary reference book to consult while answering your questions.
|
|
|
|
## How RAG Works
|
|
|
|
Think of it like this:
|
|
|
|
1. **You**: Ask a question about your documents
|
|
2. **Msty**:
|
|
- Searches your documents for relevant information
|
|
- Uses embeddings to find matches
|
|
- Selects the best chunks (default: 15)
|
|
3. **AI**:
|
|
- Receives your question and selected chunks
|
|
- Uses this default prompt:
|
|
```
|
|
The following text has been extracted from a data source due to its probable relevance to the question.
|
|
Please use the given information if it is relevant to come up with an answer and don't use anything else.
|
|
The answer should be as concise and succinct as possible to answer the question.
|
|
```
|
|
- Generates a focused answer
|
|
|
|
::image-plain{src="/how-to/create-knowledge-stack/search-settings-interface.webp" alt="Search settings showing RAG controls"}
|
|
::
|
|
|
|
::alert{type="info" icon="tabler:info-circle"}
|
|
The AI model itself never learns or remembers your documents. Each time you ask a question, Msty finds the relevant information fresh - like looking up answers in a book each time.
|
|
::
|
|
|
|
## Fine-tuning RAG
|
|
|
|
Control how RAG works in the chat interface:
|
|
|
|
1. **Similarity Threshold**:
|
|
- Low: Broader context, more results
|
|
- Medium: Balanced matching
|
|
- High: Strict matching
|
|
- Highest: Only exact matches
|
|
|
|
2. **Number of Chunks**:
|
|
- Default: 15 chunks
|
|
- More chunks = broader context
|
|
- Fewer chunks = focused answers
|
|
|
|
3. **Custom Prompt**:
|
|
- Modify the default prompt
|
|
- Guide AI response style
|
|
- Maintain answer focus
|
|
|
|
::alert{type="tip" icon="tabler:bulb"}
|
|
Pro Tip: You can select multiple Knowledge Stacks at once for cross-referencing information!
|
|
::
|
|
|
|
## Why Use RAG?
|
|
|
|
1. **Accuracy**:
|
|
- References specific facts
|
|
- Reduces "hallucinations"
|
|
- Provides sourced answers
|
|
|
|
2. **Privacy**:
|
|
- Documents stay local
|
|
- Only relevant snippets sent
|
|
- Full control over data
|
|
|
|
3. **Cost Effective**:
|
|
- Sends minimal context
|
|
- No training needed
|
|
- Works with any AI model
|
|
|
|
::alert{type="info" icon="tabler:info-circle"}
|
|
Want to optimize your results? Learn about [chunk settings](/features/knowledge-stack/advanced-features) and [embedding options](/features/knowledge-stack/embeddings).
|
|
:: |