NexusData is a local-first, agentic SQL pipeline. No cloud uploads, no SaaS subscriptions, and zero data leakage. Talk to any database natively on your hardware.
Every query undergoes a rigid anatomical breakdown before execution. We don't just "guess" SQL—we construct it through verified reasoning stages.
Sanitizes PII, standardizes casing, and removes colloquial noise using local regex + AST guards.
Deconstructs the prompt into a structured JSON "Intent Object" identifying metrics and filters.
Maps fuzzy terms (e.g., "sales") to actual DB schema (e.g., fact_orders.total_price) using LanceDB vector cache.
Determines Join paths and selects the correct SQL dialect (Postgres, Oracle, DuckDB) via SQLAlchemy 2.x.
Executes SQL. If a traceback occurs, the Self-Healer feeds the error back to the Planner for automatic correction.
Your API key is only stored in .env. It never touches our servers.
git clone https://github.com/AshwiniShankwaar/NexusData.git
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python nexus_cli.py
Your .env file is git-ignored. All database credentials remain on your disk.