Local-first RAG Application

Your documents,
finally answerable

Upload PDFs and text files, then ask questions across all of them at once. Powered by local AI — everything stays on your machine.

Get Started How It Works
localhost:8000 — DocuMind
Library · 3 docs
PDF
Research Paper.pdf
12 chunks · 84 KB
PDF
Meeting Notes.pdf
8 chunks · 32 KB
TXT
Project Spec.txt
5 chunks · 18 KB
Chat · llama3.2
Ollama online
P
What are the key findings from the research?
✦
Based on your documents, the key findings include improved model accuracy by 23%, reduced inference latency through quantization, and stronger cross-domain generalization across test sets.
P
What action items came out of the meeting?
Ask anything about your documents...
How It Works

Up and running in
five steps

DocuMind is fully local. No accounts, no API keys — just install the prerequisites and run.

01
Install Ollama & pull a model
Download Ollama from ollama.com and install it. Then run this command to download an AI model. You can use any model — llama3.2, mistral, phi3, gemma2, etc.
Terminal
ollama pull llama3.2
02
Clone the repository
Download the source code from GitHub. Requires Git installed.
Terminal
git clone https://github.com/paul-andrei-jison/RAG_document_analyzer_project.git
03
Enter the folder & install dependencies
Navigate into the project and install all Python packages. Requires Python 3.10+.
Terminal
cd RAG_document_analyzer_project
Terminal
pip install -r requirements.txt
04
Start the server
Use the first command for local use only. Use the second if you want teammates on your network to access it — then share your machine's local IP (e.g. 192.168.1.x:8000).
Terminal · local only
uvicorn main:app --reload
Terminal · open to network
uvicorn main:app --host 0.0.0.0 --port 8000
05
Open the app & start chatting
Open your browser and go to this address. Upload PDFs or TXT files using the upload button, then type any question — DocuMind searches all your documents at once to answer.
Browser address bar
http://localhost:8000
Tech Stack

Built on
solid foundations

DocuMind uses best-in-class open source tools. Everything you need for a production-grade local RAG pipeline.

⚡
FastAPI Backend & REST API
🗄
ChromaDB Vector Database
🦙
Ollama Local AI Runtime
🐍
Python 3.10+ Runtime
🌐
Vanilla HTML/CSS/JS Frontend — no framework
FAQ

Common
questions

DocuMind currently supports PDF (.pdf) and plain text (.txt) files. PDFs are parsed with PyPDF2 to extract text from each page, which is then split into chunks and embedded into ChromaDB for semantic search.
No GPU required. Ollama runs comfortably on CPU with quantized models like LLaMA 3.2. Response times will be slower without a GPU (a few seconds per reply), but it's entirely functional on most modern laptops and desktops.
Yes. Any model available through Ollama works with DocuMind. Pull a model with ollama pull <model-name>, then select it from the model dropdown in the UI. Popular options include mistral, phi3, gemma2, and qwen.
Yes. Start the server with uvicorn main:app --host 0.0.0.0 --port 8000 and share your local IP address. Anyone on the same network can access DocuMind in their browser without any additional setup.
Documents are stored locally in a ChromaDB vector database at ./local_vectordb in the project directory. The original files are deleted after ingestion — only the extracted text chunks and their embeddings are persisted. Nothing leaves your machine.