Hey HN, I built NumbyAI because I wanted an easy way to track my finances and spending without handing my financial data to a cloud service.

It's a self-hosted personal finance tool. You upload a bank statement CSV and a local LLM (Ollama, qwen3.5:9b) categorizes every transaction into 13 spending categories. A rule engine learns your preferences so repeat categorizations are instant. A dashboard gives you spending breakdowns, budget tracking, and cash flow trends across all your uploads.

Stack: FastAPI + React + Ollama + SQLite. Works on macOS, Linux, Windows.

Features:

  - Auto-detects CSV column mapping (handles EU/US date and number formats, exotic delimiters)
  - Rule engine applies saved patterns before hitting the LLM
  - Manual review queue for low-confidence categorizations
  - Dashboard with budget tracking, category breakdowns, cash flow trends
  - Rule Advisor that analyzes your patterns and suggests reusable rules

GitHub: https://github.com/RoXsaita/NumbyAI-Public Website: https://numbyai.com

Happy to answer questions about the architecture or LLM categorization pipeline.