Local Expense Sorter: an offline categorizer for freelancers and small shops

Local Expense Sorter is an offline categorization tool designed for freelancers and small business owners to manage their bank statements without uploading sensitive financial data to the cloud. The tool uses a small open-weight model running locally on the user's laptop to categorize transactions into categories such as Food & Drinks, Supplies, Rent, and Client Income.

Upon import, the tool displays monthly spend charts and exports the data into an Excel workbook or CSV file for accountants. The user can correct any categorization errors, and these corrections are saved for future reference, improving the accuracy of subsequent imports.

The categorization process prioritizes corrections made by the user, using them as examples for future transactions from the same merchant. If a merchant appears repeatedly in the same batch, the tool uses the previously corrected categorization without running a model call. The app employs Streamlit and pandas for the interface, bound to the local IP address (127.0.0.1) to ensure that it can only be accessed by the same machine.

The categorization is performed using the Qwen2.5 3B model, which is open-weight and runs locally through Ollama. For data storage, SQLite is used, ensuring that the tool is lightweight and can be easily run on any laptop. The app allows users to swap models, modify prompts, and adjust categories according to their preferences, providing greater control compared to tools that rely on cloud-based services.

However, this control comes at the cost of potential reduced accuracy due to the smaller model's capacity to handle complex bank statement descriptions.