FinanceAgent: A Personal Financial Intelligent Assistant Based on Large Language Models

FinanceAgent is an open-source personal financial intelligent assistant developed in Python. It combines local LLM (via Ollama) with bank data scraping capabilities and uses Telegram Bot as the interactive interface to provide users with intelligent financial analysis and advice. Its core goal is to bridge the gap between raw financial data and wise decision-making, with a focus on data privacy protection—all reasoning is done locally. The project was created by Alex Savizky and is open-sourced under the MIT License.

Amid the deep integration of AI and fintech, personal financial management is undergoing transformation. FinanceAgent emerged to solve the problem that raw financial data is difficult to convert into actionable investment insights, providing individual investors with an intelligent financial management tool. The project was developed by developer Alex Savizky and is open-sourced under the MIT License, allowing free use and modification.

FinanceAgent adopts a modular architecture, with core modules including:

• brain.py: The intelligent brain, which interacts with Ollama's local LLM, constructs prompts, and generates responses. It uses telegramify_markdown to ensure proper display on Telegram.
• main.py: The Telegram Bot interface, based on the telebot library, implementing command handling, conversation management, and whitelist access control.
• bank_connector.py: The bank data connector, currently supporting Bank Leumi (Israel's National Bank), which scrapes account information via Node.js scripts.
• database.py: The data persistence module, storing user profiles and conversation history.
• scraper/: The data scraping layer written in Node.js, fetching real-time bank data.

The tech stack mixes Python (for AI reasoning and business logic) and Node.js (for web scraping). Each module has clear responsibilities and is easy to extend.

FinanceAgent has the following core features:

1. Personalized User Profile: Collects user information such as name, age, and monthly income through guided conversations, providing a basis for targeted advice.
2. Real-time Bank Data Integration: Connects to Bank Leumi to obtain real-time account balances and transaction records, identifying large expenditures without manual data entry.
3. Conversational Financial Analysis: Supports natural language interaction, maintains conversation context, generates responses based on user profiles and history, and uses local LLM to ensure privacy.
4. Secure Access Control: A whitelist mechanism based on Telegram user IDs ensures only authorized users can access sensitive financial information.

Choosing Ollama as the local LLM backend has the following advantages over cloud APIs:

• Data Privacy: User financial data (bank records, transaction information) does not leave the local machine, eliminating the risk of data leakage.
• Cost Control: No API costs based on token usage; only hardware and electricity costs for local operation are incurred.
• Offline Availability: Except for bank data scraping, the system can run offline, adapting to scenarios with unstable networks.
• Model Flexibility: Users can choose models of different scales (e.g., Phi-3, Llama 3 series) based on their hardware conditions.

Target Users: Tech enthusiasts, privacy-sensitive users, Bank Leumi customers, self-hosting lovers. Deployment Process:

1. Clone the code repository; 2. Configure the Ollama local LLM service; 3. Set up the Telegram Bot Token; 4. Configure bank access credentials; 5. Start the service. Expansion Potential: Support for other banks can be added by new bank_connector modules; the analysis style and advice types can be customized by modifying the prompts in brain.py.

The current project has the following limitations and improvement directions:

• Limited Bank Support: Only supports Bank Leumi; more bank connectors need to be developed.
• Error Handling to Be Improved: Optimize the user experience when bank data scraping fails, providing detailed error information and recovery suggestions.
• Basic Features: Advanced features such as investment analysis, budget planning, and financial goal tracking can be expanded.
• Lack of Visualization: Add chart components to enhance the presentation of analysis results.

FinanceAgent demonstrates the application potential of AI technology in personal financial management. By combining local LLM with real-time financial data, it provides an intelligent and privacy-friendly prototype of a financial assistant. With the advancement of local LLM technology and the maturity of the open-source financial API ecosystem, such tools are expected to become more powerful and user-friendly. This project provides a valuable reference implementation for the AI+finance field and is worth the attention and exploration of developers.