FinanceBro - Stock Analysis & Portfolio Bot

A Telegram bot built on Cloudflare Workers for automated financial analysis and portfolio management. Uses artificial intelligence (Anthropic Claude) to generate investment reports and interpret trading commands, integrating real-time market data from Financial Modeling Prep (FMP).

Architecture

The project is developed in Node.js and deployed on the Cloudflare ecosystem:

• Cloudflare Workers: Serverless handling of Telegram webhook requests.
• Durable Objects: Persistent state management (SQLite), background queues (Alarms), and user data consistency.
• External Integrations:
  • Telegram Bot API: Main user interface.
  • Anthropic API (Claude 3.5 Sonnet): Financial analysis engine and natural language parsing.
  • Financial Modeling Prep (FMP) API: Data source for real-time quotes, price history, and company profiles.

Features

1. AI Financial Analysis

The /analyze [TICKER] command initiates a multi-step analysis process:

1. The Worker delegates the task to the Durable Object via the Alarm system to handle long-running executions.
2. The AI collects market data (price, changes, market cap, P/E) and historical data (30-day OHLC) via Tool Use.
3. A structured report is generated including:
   • Fundamental analysis (Business model, financial health).
   • Technical analysis (Trend, support/resistance levels).
   • Weighted score (Technical 35%, Fundamentals 30%, Sector 20%, Sentiment 15%).
   • Operational recommendation (BUY/PASS/WATCHLIST) with execution strategy and risk management.

2. Portfolio Management and Trading

The system tracks positions and calculates P&L in real-time.

• Order Entry: Supports structured commands (/buy AAPL 150 10) and natural language (I bought 10 Apple shares at 150 dollars). The AI performs parsing if pattern matching fails.
• Portfolio Tracker: The /portfolio command returns a table with open positions, calculating profit/loss based on current market prices retrieved from FMP.
• Budgeting: The /setbudget command sets available capital, influencing the AI's position sizing suggestions.

3. Security and Rate Limiting

• Authentication: Optional support for access password (BOT_PASSWORD) to restrict bot usage.
• Rate Limiting: SQLite-based system within the Durable Object to limit daily AI analyses per user.

Configuration

The project requires the following environment variables (configurable in wrangler.toml or Cloudflare dashboard):

Variable	Description
TELEGRAM_BOT_TOKEN	Token obtained from @BotFather
ANTHROPIC_API_KEY	API key for accessing Claude models
FMP_API_KEY	Financial Modeling Prep API key
BOT_PASSWORD	(Optional) Password to protect bot access

Development and Deployment

Prerequisites

• Node.js
• Cloudflare account

Install Dependencies

npm install

Deploy

Deployment to Cloudflare Workers is done via Wrangler:

npx wrangler deploy

Code Structure

• src/index.js: Main router and webhook handling.
• src/user-storage.js: Durable Object implementation. Manages SQLite, background queues, and persistent business logic.
• src/ai.js: Interface with Anthropic API. Handles prompt engineering, response streaming, and tool execution.
• src/fmp.js: HTTP client for Financial Modeling Prep APIs.
• src/telegram.js: Utilities for message formatting and Telegram API calls.