GraphMind: Terminal Hybrid Knowledge-Graph & Vector RAG

GraphMind is a local terminal-based retrieval system for asking questions over your own knowledge base. It ingests .pdf, .txt, or .md files, builds a knowledge graph plus FAISS/BM25 indexes, and answers questions through a hybrid retriever.

Architecture

flowchart LR
    A["Source file (.pdf/.txt/.md)"] --> B["Ingestion"]
    B --> C["spaCy NER + Triplets"]
    C --> D["Entity Resolution"]
    D --> E["NetworkX Graph"]
    B --> F["SentenceTransformer Embeddings"]
    F --> G["FAISS IndexFlatIP"]
    B --> H["BM25 Index"]
    E --> I["Graph Retrieval"]
    G --> J["Vector Retrieval"]
    H --> K["Keyword Retrieval"]
    I --> L["Hybrid Scorer"]
    J --> L
    K --> L
    L --> M["LLM Provider"]
    M --> N["Terminal Answer"]

Loading

Features

Capability	Implementation
Source ingestion	Supports .pdf, .txt, and .md.
Chunking	Deterministic sentence-window chunking with overlap.
Entity deduplication	Canonical graph node resolution using string similarity.
Hybrid retrieval	Vector search, BM25 keyword search, and graph traversal.
Concurrent retrieval	Retrieval paths run in parallel with ThreadPoolExecutor.
LLM providers	ollama, openai, or gemini via .env.

Setup

python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
python -m spacy download en_core_web_sm
copy .env.example .env

For local Ollama:

ollama pull llama3.2

Configure LLM

Edit .env.

For Ollama:

LLM_PROVIDER=ollama
LLM_MODEL=llama3.2

For Gemini:

LLM_PROVIDER=gemini
LLM_MODEL=gemini-3.5-flash
GEMINI_API_KEY=your_key_here

For OpenAI:

LLM_PROVIDER=openai
LLM_MODEL=gpt-4o-mini
OPENAI_API_KEY=your_key_here

Add Your Knowledge Base

Put your source files inside knowledge_base/:

knowledge_base/
  notes.pdf
  notes.txt
  notes.md

Then build the artifacts:

python ingestion.py
python graph_engine.py
python vector_engine.py

To ingest a specific file or another folder:

python ingestion.py --input path\to\file.pdf
python ingestion.py --input path\to\folder

Generated files are written to data/:

Artifact	Purpose
chunks.pkl	Chunk text and entity metadata.
triplets.pkl	Extracted subject-predicate-object facts.
graph.pkl	NetworkX graph.
faiss_index.bin	Vector index.
chunk_map.pkl	FAISS ID to chunk ID map.
inverted_index.pkl	BM25 keyword index.

Ask Questions

python demo.py

Type your question and press Enter. Press Enter on a blank prompt to exit.

Repository Layout

graphrag/
  demo.py             Terminal question-answering interface
  ingestion.py        Source loading, chunking, NER, triplet extraction
  graph_engine.py     Entity resolution and knowledge graph construction
  vector_engine.py    FAISS vector index and BM25 index build
  retriever.py        Concurrent hybrid retrieval and LLM provider routing
  neo4j_export.py     Optional graph export to Neo4j
  knowledge_base/     Put local .pdf/.txt/.md files here
  requirements.txt    Runtime dependencies
  .env.example        Configuration template