A private, local AI assistant for your documents—without sending data to the cloud

Users upload their PDF, DOCX, TXT, or notes (e.g., from Obsidian, Notion, or personal files) and receive a local AI assistant that:

• Answers questions about document content.
• Finds citations, summarizes sections, and compares files.
• Works completely offline on their computer (Mac, Windows, or Linux).
• No data is transmitted online—maximum privacy.

• 📄 Document upload: PDF, TXT, Markdown

• 💬 AI-powered chat based on your documents (RAG)

• 🕵️‍♂️ Complete privacy – everything runs on your computer

• 🧠 Uses local LLM via Ollama (phi3, Mistral, Llama 3, etc.)

• 📜 Chat history is saved

• 🌐 Simple web interface (or desktop app)

  🔧 Technologies

• Language: Python (base) + Electron or Tauri (for GUI)

• LLM: Ollama (phi3, Mistral, Llama 3)

• Embeddings + RAG: ChromaDB or FAISS

• Frontend: React + Vite (if using Tauri) or pure HTML/CSS for simplicity

• Documents: PyPDF2, built-in parsers

  🔧 Requirements

• Ollama (install and run)

• Python 3.9+

• pip

  💰 Monetization

• Free version (open-source):

• Supports basic formats (TXT, PDF)

• Works with small files

• CLI interface

  Paid "Pro" version (SaaS or desktop license)

• Support for DOCX, PPTX, Excel, and Notion exports

• Improved UI with chat history and tags

• Syncing across devices (optional, with encryption)

• Priority support and updates

• Selling through Gumroad or your own website (~$15–$29/time or $5/month)

  Additionally:

• GitHub Sponsors to support development

• Partnerships with privacy software vendors (Proton, Tuta, etc.)

  🎯 Target audience

• Lawyers, doctors, and researchers who work with confidential documents.

• Product managers and analysts who analyze internal reports.

• Regular users who are tired of ChatGPT "remembering" their data.

  🌍 Why is this relevant?

• Growing interest in local LLMs (Ollama, LM Studio, Jan.ai).

• Privacy concerns when using ChatGPT.

• Many want an AI assistant but don't want to be dependent on the cloud.

  🗺️ 1. Project Architecture (LocalDocs AI)

┌───────────────────────────────────────────────────────┐
│                  LocalDocs AI (Desktop App)           │
└───────────────────────────────────────────────────────┘
                             │
         ┌───────────────────┼───────────────────┐
         │                   │                   │
┌─────────▼─────┐   ┌─────────▼─────┐   ┌─────────▼─────┐
│   Frontend    │   │   Backend     │   │   AI Engine   │
│  (Tauri/React)│   │ (FastAPI/Flask)│  │ (Ollama/Llama.cpp)│
└───────────────┘   └───────┬───────┘   └───────▲───────┘
                           │                   │
                   ┌───────▼───────┐   ┌───────┴───────┐
                   │  Document     │   │  Vector       │
                   │  Storage      │   │  Database     │
                   │  (./docs/)    │   │  (ChromaDB)   │
                   └───────────────┘   └───────────────┘

🔁 Data flow:

• The user adds a PDF → saves it to ./docs/.
• The system parses the text and breaks it into chunks.
• The chunks are embedded and saved in ChromaDB.
• When asked: → the query is embedded → relevant chunks are searched → → a prompt is generated → sent to the local LLM → → a response is returned

LocalDocs AI
│
├── core/
│   ├── document_loader.py   ← support PDF, TXT, MD
│   ├── rag_engine.py        ← RAG + chat history
│   └── chat_history.py      ← history management
│
├── api/
│   └── main.py              ← FastAPI server
│
├── static/                  ← simple HTML interface
│   └── index.html
│
├── docs/                    ← user files
└── chroma_db/               ← vector database

Install:

pip install -r requirements.txt

And make sure Ollama is running and the model is loaded:

ollama pull phi3  # mistral, llama3

• Create a docs/ folder and place any PDF there (name it sample.pdf).
• Launch Ollama: ollama serve (it usually starts automatically).
• Run:

python main.py add docs/sample.pdf
python main.py ask "What is this document about?"

git clone https://github.com/mscbuild/local-docs-ai-agent.git
cd local-docs-ai-agent
pip install -r requirements.txt

cd api
uvicorn main:app --reload --port 8000

Open in your browser: http://localhost:8000

Usage

• Upload a PDF/TXT/MD file
• Ask questions: "What is this document about?", "Find the contract date," etc.
• Get answers from AI trained on your data

• All documents are stored locally (./docs/)
• The vector database is on your disk (./chroma_db/)
• Requests to LLM are processed through Ollama on your machine
• Not a single byte is lost to the internet

This project is open source. If you find it useful:

• ⭐ Star it on GitHub
• 💬 Report bugs in Issues
• 💰 Support the developer through GitHub Sponsors (coming soon)

License – free to use for personal and commercial purposes.