Add Bloomberg terminal clone with Claude integration

[SecretPoets]

Description

This PR adds a complete Bloomberg-style terminal web application with Claude AI integration. The application demonstrates multiple Claude API capabilities in a single, cohesive demo.

What it includes

Frontend (frontend/):

• index.html: Multi-panel Bloomberg terminal layout with watchlist, quote detail, news feed, and chat pane
• style.css: 643 lines of Bloomberg-inspired styling (black background, amber text, CRT aesthetic)
• app.js: Vanilla JavaScript frontend that fetches market data and streams chat responses

Backend (backend/):

• main.py: FastAPI server serving static files and REST/SSE endpoints
• mock_data.py: Self-contained mock financial data (quotes, history, news, indices, FX, commodities, crypto)
• claude_client.py: Claude integration showcasing four API features:
  • Tool use: Claude calls get_quote, get_history, get_news, market_overview tools
  • Streaming: Tokens pushed to frontend as Server-Sent Events
  • Extended thinking: Optional thinking budget with separate UI panel
  • Prompt caching: Long system prompt cached with cache_control: ephemeral

Launcher & Dependencies:

• run.py: Entry point that loads .env and starts uvicorn
• requirements.txt: FastAPI, uvicorn, anthropic, pydantic, python-dotenv

Key features

• No build step: Vanilla JS, no npm/webpack required
• Offline-capable: All market data is mock data, no real API calls
• Agentic loop: Tool-call → tool-result loop in stream_chat() until Claude stops requesting tools
• Real-time UI: Watchlist and ticker strip update every 8-10 seconds; chart renders on symbol selection
• Command interface: Type ticker symbols, NEWS, CHAT <question>, or HELP in the command bar
• Responsive layout: Grid adapts from 3-column to 2-column on narrow screens

Why this is valuable

This cookbook demonstrates how to build a real-world, multi-feature Claude application with:

• Proper tool definition and dispatch
• Streaming integration with a web frontend
• Prompt caching for cost efficiency on repeated queries
• Extended thinking for complex financial analysis
• Clean separation of concerns (frontend/backend/mock data)

All in a visually polished, immediately recognizable interface that makes the capabilities tangible.

Type of Change

• New cookbook
• Bug fix
• Documentation update
• Code quality improvement
• Dependency update
• Other

Cookbook Checklist

• Cookbook has a clear, descriptive title ("TERM-AI: Bloomberg Terminal Clone")
• Includes a problem statement (building a multi-feature Claude app with tool use, streaming, thinking, caching)
• Code is well-commented (docstrings in Python, inline comments in JS)
• Includes expected outputs (README describes layout and feature matrix)
• Added entry to registry.yaml (to be done in follow-up or by maintainers)

Testing

• Tested locally: frontend loads, watchlist/quote/news panels populate from mock data, chat streams responses with tool calls
• All Python files are syntactically valid and follow project conventions
• Frontend HTML/CSS/JS are valid and responsive
• No external API calls required (mock data only)

Additional Context

How to run:

cd bloomberg-terminal-clone
pip install -r requirements.txt
cp .env.example .env  # Add ANTHROPIC_API_KEY
python run.py
# Open http://127.0.0.1:8000

Model used: claude-sonnet-4-6 (per project conventions)

Files added: 11 (HTML, CSS, JS, Python backend, mock data, launcher, requirements, README)

This is a self-contained, runnable demo that can serve as a reference for building production Claude applications with tool use, streaming, and advanced features.

https://claude.ai/code/session_01F6i5AgvYsgLnMvUYRk7ELQ