Independent Reviewer

FinAlly — AI Trading Workstation

A dark-themed, Bloomberg-inspired trading terminal with live-streaming market data, simulated portfolio management, and an AI chat assistant that can analyze positions and execute trades via natural language.

Built entirely by coding agents as the capstone for an agentic AI coding course.

Quick Start

cp .env.example .env   # add your DATABASE_URL and OPENROUTER_API_KEY
docker compose up       # open http://localhost:3000

A default user is pre-seeded with $10,000 in virtual cash and a watchlist of 10 tickers. No login required.

Architecture

Layer	Stack
Frontend	Next.js, TypeScript, Tailwind CSS, Lightweight Charts
Backend	FastAPI, Python (uv), LiteLLM
Database	Neon Serverless Postgres
Real-time	Server-Sent Events (SSE)
AI	LiteLLM → OpenRouter (Cerebras inference)

Two Docker containers orchestrated via docker-compose.yml. The frontend proxies /api/* to the backend through Next.js rewrites — single port, no CORS.

Features

• Live price streaming — prices flash green/red on change with sparkline mini-charts
• Simulated trading — market orders, instant fill, fractional shares
• Portfolio heatmap — treemap sized by weight, colored by P&L
• P&L tracking — portfolio value chart from periodic snapshots
• AI assistant — analyzes positions, suggests and auto-executes trades, manages watchlist

Environment Variables

Variable	Required	Description
DATABASE_URL	Yes	Neon Postgres connection string
OPENROUTER_API_KEY	Yes	OpenRouter API key for AI chat
MASSIVE_API_KEY	No	Massive API key for real market data (simulator used if absent)
LLM_MOCK	No	Set true for deterministic mock LLM responses

Project Structure

frontend/    # Next.js app (port 3000)
backend/     # FastAPI app (port 8000)
planning/    # Project spec and agent docs
test/        # Playwright E2E tests

Documentation

See planning/PLAN.md for the full project specification.

License

See LICENSE.