Managing Knowledge Base

What is Memories?

Memories is a local knowledge base feature in Sapientia that stores documents and important information into a vector database. This data is used to answer your questions more personally and relevantly using semantic search techniques.

Benefits:

• Data is stored entirely on your local device
• AI can answer based on private/internal documents
• Smart search using embedding similarity

---

How to Upload Files to Memories

1. Click the "Insert" button on the Memories page
2. Select one or multiple files (PDF/DOCX)
3. Progress is displayed in real-time (number of chunks processed)
4. Complete → files ready to use for RAG

Notes:

• Upload can only be performed when LLM is already running (requires embedding model)
• Large files will be split into multiple chunks according to file size

---

Filter Memories

Filter Query:

• Input: source = 'C:\Users\Admin\notes.pdf'
• Uses WHERE clause to filter vector database

Search Fragment (Semantic Search):

• Input: free text such as "project documentation"
• System converts to embedding → finds most similar chunks

---

How Memories are Used in RAG

When you ask something:

1. Your query is converted into an embedding vector
2. System finds top-5 most relevant chunks
3. Those chunks are sent as context to AI:

   --- CONTEXT (Use ONLY the information below...) ---
   [Fragment 1], Source: file.pdf
   [Fragment 2], Source: document.docx
   --- END OF CONTEXT ---

4. AI answers based on that context

Example:

• User: "What's in notes.pdf about the meeting?"
• System: Find chunks from notes.pdf → send as context → AI answers