Personal Gmail RAG System

Local RAG pipeline that ingests Gmail emails, stores them in a vector database, and supports querying via an open-source LLM. Keeps previous email context for follow-up questions.

Install

python setup.py

Activate

source venv/bin/activate

Credentials

1. Go to Google Cloud Console
2. Search API & Services
3. Open Credentials
4. Click Create Credentials → OAuth Client ID
5. Select Desktop app
6. Download the JSON file
7. Move it to src/credentials.json

Enable the Gmail API.

Ollama Model

Install the required Mistral model:

ollama pull mistral

The default mistral model uses Q4_K_M quantization (4.4GB size vs 14GB FP32) for optimal performance.

Database

./setup_database.sh

Usage

python src/main.py

Benchmark

python benchmark.py

Demo

Important

Please follow previous steps to setup the enviornment!

Run demo

python demo.py

References

• Gmail API Quickstart
• Gmail Search Operators
• Gmail API Message Format
• Gmail API attachments Format
• Email Header Specifications
• Refine searches in Gmail
• Installing PostgreSQL + pgvector on Debian
• pypdf docs
• Mime types
• Improving Performance of PostgreSQL queries
• pytesseract project desc
• Vector Similarity Search with PostgreSQL’s pgvector – A Deep Dive
• Introduction: From Distance to Cosine Similarity
• Unlocking the Power of Sentence Embeddings with all-MiniLM-L6-v2
• Mistral-7B
• LangChain