Project

Accounty

A local-first accounting assistant with Streamlit UI, FinTS banking, Paperless-ngx archiving, and AI-supported invoice workflows.

Accounty shows how AI-assisted accounting can be implemented pragmatically: not as a vague copilot concept, but as a concrete local workflow that brings together banking, document handling, and user confirmations in one interface. The source project is available on GitHub.

Project summary

Accounty is an open-source accounting agent for small businesses. The application combines a Streamlit chat interface with MCP-based tools for invoice handling, SEPA payment preparation via FinTS/HBCI, Paperless-ngx archiving, and filesystem storage, while keeping deployment local for stronger data privacy.

Challenge

  • Small teams often handle invoices, bank transactions, and document archiving across separate tools, which creates manual effort and process breaks.
  • Accounting automation still needs practical approval steps, TAN-based banking interactions, and clear user control instead of opaque black-box behavior.
  • The project needed a foundation that is useful today for core workflows and still extensible for later capabilities such as email ingestion, local LLMs, and broader document search.

Solution

  • Built a local-first web app in Streamlit with a chat-driven workflow and quick actions for recurring accounting tasks.
  • Connected dedicated MCP servers for FinTS banking, Paperless-ngx, document processing, and filesystem operations so the agent can work through bounded tools.
  • Used LangGraph, LangChain MCP adapters, and OpenAI-based prompting to combine invoice extraction, transaction context, SEPA draft execution, and document archiving in one flow.

Results

  • A usable early-stage accounting assistant that can already read invoices, show balances and recent transactions, create SEPA transfer flows with TAN input, and archive documents.
  • A privacy-conscious deployment model: the application is designed to run locally, with sensitive banking and document data staying in the user's own environment.
  • A clean extension path for next features such as inbox ingestion, local model support, richer Paperless metadata, and Docker-based packaging.