RAG in a chat question

[PixeroJan]

Ok, I’m trying to understand how to use RAG within a chat. I have some data that I want the chat to keep in mind when asking it to do stuff. I don’t want it to just search for that data but have it as background info for the chat. Until now I’ve attached some docs when chatting but it takes a lot of tokens so I started to look into RAG. I’m a Copilot Plus user and have tried to find info how to do this but I can’t understand how. Is it even possible? I mostly use Gemini 2.5 pro or Claude 3.7 Sonnet for these tasks.

[ichts]

The plugin uses a hybrid RAG system (Search v3) that combines:

1. Lexical Search (always active, no index needed)
   - Fast, memory-bounded keyword search
   - Chunk-based retrieval from your notes
   - Instant setup, no AI model required
2. Semantic Search (optional, AI-powered)
   - Vector embeddings for similarity matching
   - Understands meaning, not just keywords
   - Uses embedding models (OpenAI, Cohere, etc.)
   - Stored in Orama vector database

When enabled together: Merged retrieval that combines both approaches for best results.

---

How to Set It Up

Step 1: Enable Semantic Search

1. Open Obsidian Settings → Copilot
2. Go to QA section
3. Toggle "Enable Semantic Search"
4. Choose an Embedding Model: since you are using copilot plus, you can choose copilot-plus-small or copilot-plus-multilingual

Step 2: Configure Indexing

In the QA settings:

• Auto-Index Strategy: Set to ON MODE SWITCH (recommended)
  • This indexes your vault when you first open Vault QA mode
  • More efficient than indexing on every startup

Step 3: Choose What to Index

• Exclusions (qaExclusions): Add paths or tags to exclude
  • Example: templates/, daily/, #private
  • One per line
• Inclusions (qaInclusions): Only index specific folders
  • Leave empty to index entire vault
  • Example: research/, projects/

Step 4: Index Your Vault

Option A: Automatic

• Switch to Vault QA mode
• Click "Yes" when prompted to index
• Wait for indexing to complete (1-2 minutes for typical vaults)

Option B: Manual Commands

• Press Ctrl/Cmd+P → "Refresh Index for Vault"
• Or "Force Reindex Vault" for full rebuild

---

How It Works in Practice

1. Vault QA Mode

• Best RAG experience
• Use chain type: vault_qa or copilot_plus
• AI automatically searches your indexed notes
• References appear in chat (clickable links to source notes)

2. Regular Chat Mode

• Works but with limited RAG context
• Only searches a few recent notes
• Better to use Vault QA for knowledge-heavy questions

3. Example Workflow

You: "Summarize the key insights from my project planning notes"
Plugin:
1. Searches indexed notes for "project planning"
2. Finds 5 relevant chunks across 3 notes
3. Provides 5 sources with links
4. AI generates summary using that context
5. Response includes: "Based on your notes [Link 1] [Link 2]..."

[PixeroJan]

Thank you for your detailed answer.
So no good way of chatting with RAG at the moment.
I´ll have to think about how I can find a way of achieving this another way.
Would it be possible to have another plugin creating vectorized chunks and feeding that to copilot?

[logancyang]

wdym? Vault search is RAG.

[PixeroJan]

Maybe I'm not using the correct terms. I want to chat using Gemini 2.5 pro or Claude 3.7 Sonnet but I want copilot to know facts from my vault while chatting. Now I use [[filename]] for all files but that uses LOTS of tokens so I'm searching for an alternative approach and my idea was that using RAG for that info the copilot chat could find the "headlines" and retrieve just the needed info without me having to manually add everything. Could it be achieved some other way? What if I had a "header" document with resumes and links to the actual files? Would the chat still "download it all" or just what is needed for that perticular task.

[ichts]

Vault QA is made for this. Have you tried?

[PixeroJan]

Vault QA is asking questions about files in the vault. That is not the same as having the chat have background info that it can use while completing tasks.