========================================================================
██████╗ ██████╗ ██████╗███╗ ███╗██╗███╗ ██╗██████╗ █████╗ ██╗
██╔══██╗██╔═══██╗██╔════╝████╗ ████║██║████╗ ██║██╔══██╗ ██╔══██╗██║
██║ ██║██║ ██║██║ ██╔████╔██║██║██╔██╗ ██║██║ ██║ ███████║██║
██║ ██║██║ ██║██║ ██║╚██╔╝██║██║██║╚██╗██║██║ ██║ ██╔══██║██║
██████╔╝╚██████╔╝╚██████╗██║ ╚═╝ ██║██║██║ ╚████║██████╔╝ ██║ ██║██║
╚═════╝ ╚═════╝ ╚═════╝╚═╝ ╚═╝╚═╝╚═╝ ╚═══╝╚═════╝ ╚═╝ ╚═╝╚═╝
========================================================================
INTELLIGENT LOCAL & CLOUD RAG DESKTOP
DocMind AI is an intelligent desktop application that lets you run local and cloud-based Retrieval-Augmented Generation (RAG) queries against your document libraries. Built entirely in Python using CustomTkinter for a premium responsive desktop interface, and LangChain + FAISS for search indexing, it represents a clean, modular assistant with advanced controls and detailed session analytics.
The application splits concerns across an Object-Oriented design separating UI views, DB management, model generation, and index retrieval.
graph TD
User([User]) <--> GUI[CustomTkinter App Interface]
GUI <--> DB[(SQLite Database)]
GUI <--> RAG[RAG Pipeline Controller]
subgraph Ingestion Pipeline
Doc[Source Files PDF/Word/Text] --> Proc[Document Processor]
Proc --> Spl[Recursive Chunk Splitter]
Spl --> Embed[Embedding Manager]
Embed --> FAISS[(Local FAISS Indexes)]
end
subgraph Query Execution
RAG --> Retrieval[Merged FAISS Retrieval]
Retrieval --> Prompt[Context prompt builder]
Prompt --> LLM[LLM Manager]
LLM --> Stream[Callback token streaming]
Stream --> GUI
end
All major computations (such as document indexing and LLM execution) run asynchronously in dedicated background threads to prevent UI freezes.
sequenceDiagram
autonumber
actor User
participant GUI as App GUI Thread
participant Worker as Background Ingestion Thread
participant DB as SQLite DB
participant FAISS as FAISS Vectors
User->>GUI: Ingest Documents (Drag / Browse)
GUI->>Worker: Spin ingestion worker thread
activate Worker
Worker->>DB: Add document metadata entry
Worker->>Worker: Parse text page-by-page
Worker->>Worker: Split into character chunks
Worker->>DB: Update pages and chunks stats
Worker->>FAISS: Index chunks & build local FAISS store
Worker-->>GUI: Update progress bars and completion
deactivate Worker
GUI->>User: Display Indexed files list with stats
- Real-time library stats: total document count, total parsed vector chunks, active LLM model engines, storage footprints, and engagement metrics.
- Unified Global Semantic Search looking up matching filenames, message history logs, and document snippets concurrently.
- Quick Actions bar to jump instantly to settings, uploads, or AI chat tabs.
- View comprehensive document stats (file size, uploaded timestamp, parsing statuses, page count, and chunk splits).
- Perform individual actions: Open (with native OS default app), Index (re-index using new chunk constraints), Meta (inspect metadata card), Rename, and Delete.
- Unified AI Insights generation providing an executive summary, core keywords list, and suggested follow-up questions for any document.
- Real-time token-by-token streaming with interactive Stop Generation control.
- Citation drawer for each response showing matching document page numbers, source texts, and similarity scores.
- Clickable citation headers that launch a Built-in PDF Page Viewer rendering the exact page source as a high-fidelity image or text block.
- Export chat session transcripts to plain TXT or styled PDF.
- Real-time aggregated statistics from SQLite DB.
- Automatically calculates and visualizes Most Asked Topics using keyword frequency algorithms and Most Cited Documents based on history citations.
- Tracks and displays real query response latencies instead of mock benchmarks.
- Python
- CustomTkinter
- LangChain
- FAISS
- SQLite
- Sentence Transformers
- PyMuPDF
- python-docx
- Pillow
- Ollama
- Groq API
- OpenAI API
DocMind-AI/
├── app.py # Main application entry point
├── database.py # SQLite database operations
├── document_processor.py # PDF, DOCX and TXT processing
├── embeddings.py # Embedding model management
├── vector_store.py # FAISS vector index management
├── llm_manager.py # LLM providers (Ollama, Groq, OpenAI)
├── chat_manager.py # Chat session handling
├── rag_pipeline.py # Retrieval-Augmented Generation pipeline
├── utils.py # Utility functions
├── requirements.txt # Project dependencies
├── README.md # Project documentation
├── database/ # Generated automatically at runtime
├── documents/ # User uploaded documents
├── vectors/ # Generated FAISS indexes
├── screenshots/ # Application screenshots
└── ui/
├── components.py
├── login.py
├── dashboard.py
├── chat.py
├── upload.py
├── settings.py
├── history.py
├── analytics.py
└── viewer.py
git clone https://github.com/minbot616/DocMind-AI.git
cd DocMind-AIpython -m venv venvWindows
venv\Scripts\activatemacOS/Linux
source venv/bin/activatepip install -r requirements.txtpython app.pyInstall Ollama and download a model:
ollama pull llama3Start the Ollama service before launching the application.
All options can be updated directly within the Settings & Preferences tab:
- Appearance: Switch theme modes dynamically (Dark, Light, System).
- Embeddings Settings: Select indexing embeddings models (
all-MiniLM-L6-v2,all-mpnet-base-v2, oropenai). - Text Chunking: Customize Chunk Size and Chunk Overlap values to optimize search granularity.
- AI Provider: Configure LLM providers (Ollama, Groq, OpenAI) and input API keys.
- Model Settings: Configure active model, generation Temperature, and Maximum Tokens.
- Connection Tester: Click Test LLM Connection to query selected endpoints asynchronously without freezing the interface.
- BM25 Hybrid Retrieval: Combine keyword lookups with vector search for better keyword matching.
- Reranking: Add flashrank or cohere rerankers to narrow down high-relevance chunks.
- OCR Ingestion: Integrate Tesseract OCR libraries to process scanned assets.
- GPU Acceleration: Detect CUDA or MPS platforms automatically to load sentence-transformers faster.