Getting Started with NumbyAI

NumbyAI is a personal finance categorizer with a strong stance on data control. It parses bank CSVs, detects statement structure automatically, and categorizes transactions using an LLM when needed. It is not a SaaS dashboard that asks for sensitive data — it is built around local processing.

Three layers working together: Upload and parse. Categorize with rules, then AI fallback. Review and improve, then reuse what you learn.

Who this is for

This is useful if you get statements from multiple banks and hate manual remapping, want transparent logic for categories, prefer local tools, and want to keep CSV content out of cloud pipelines.

Getting started

1. 1.Clone the repository: git clone https://github.com/RoXsaita/NumbyAI-Public.git
2. 2.Install Ollama and pull the default model: python run.py setup-ollama
3. 3.Copy the environment file: cp server/.env.example server/.env
4. 4.Start everything: python run.py start

The default URL is http://localhost:8000.

Architecture

• •Frontend at port 8000 with upload, review, and dashboard views
• •FastAPI backend handles statement analysis, rule matching, and task orchestration
• •Ollama handles category inference for uncertain rows
• •SQLite locally by default, PostgreSQL available for production
• •Optional auth integration, but single-user local mode works without it

Configuration

Check server/.env before first run: - OLLAMA_URL: keep http://localhost:11434 - OLLAMA_MODEL: default qwen3.5:9b - CATEGORIZATION_BATCH_SIZE: tune for throughput - DATABASE_URL: keep SQLite for local use

If your first run feels slow, spend ten minutes reviewing low-confidence rows and creating rules. Next import usually gets faster.