DataMind is an all-in-one intelligent assistant built on LlamaIndex, integrating five core modules:

Module	Description	Backend
RAG	Semantic vector retrieval with multimodal support (CLIP / VLM textualization)	Chroma
GraphRAG	Knowledge graph retrieval	NetworkX
Database	Natural language to SQL	SQLite
Skills	Extensible tool system	FunctionTool
Memory	Conversation memory	Short-term + Long-term

The Agent automatically selects the right tool based on the user's question — no manual specification needed.

Why DataMind?

The name captures exactly what the system does: Data is static raw material; Mind brings it to life through understanding, reasoning, memory, and decision-making.

Each module maps to a cognitive capability:

• RAG → Perception (encoding raw text into vector representations)
• GraphRAG → Association (linking concepts through entity-relation networks)
• Database → Language (bridging natural language and formal query languages)
• Memory → Memory (working memory + long-term memory with automatic summarization)
• Skills → Skills (procedural knowledge — learned abilities the system can invoke)
• Agent → Executive function (autonomously deciding which capability to use)

Two Ways to Use

• Web UI — Full-featured interface with real-time streaming, document management panels for RAG/GraphRAG/Database/Skills/Memory
• Terminal CLI — Interactive command-line mode, same functionality, suitable for headless servers

Design Goals

1. Modular & Extensible — Each module is independent; add new retrieval strategies or tools without touching others
2. Benchmark-Ready — Built-in concurrent inference benchmarking with accuracy evaluation
3. Data-Centric — Profile-based data management for comparing different preprocessing strategies
4. Zero GPU Required — All LLM inference and embedding via remote API