Roadmap
Current state: v4.0
Money OS is a financial AI co-pilot with 20 skills, a market screening data service, paper trading, and a backtest engine — all running against real market data.
Completed milestones
v3.0–3.1: Foundation (shipped)
- 17 financial skills across 5 layers (cash flow, portfolio, tax, wealth, emotional)
- Profile persistence in local markdown files
- Unified
/money-os intent router
- Zero-trust local-only architecture
v4.0: Market Intelligence + Autonomous Trading Foundation (shipped)
- Security Screener data service — Next.js 15, TypeScript, Supabase PostgreSQL
- 4 technical indicators — ATR(14), RSI(14), MACD(12,26,9), Zigzag pivot detection
- 3 computation engines — trendline detection, VIX-adjusted scanner, signal generator
- Real market data — 110 tickers via Yahoo Finance, daily + weekly OHLCV
- Backtest engine — walk-forward simulation with parameter sweep optimization
- Paper trading — virtual $100K portfolio with stop-loss/take-profit, trade proposals
- Trade gate — human approval pipeline with configurable auto-approve rules
- Investment Navigator — GPS-style
/invest command (say a goal, get a step-by-step path)
- ADEPT coaching — Analogy → Diagram → Example → Plain interpretation → Technical abstraction
- 53 passing tests (unit + functional)
- 12 API endpoints serving real data
PRDs:
Upcoming milestones
v4.1: Expanded Coverage + Smarter Signals
Goal: more asset classes, fundamental data, and a regime filter that stops buying bounces in bear markets.
Deliverables:
- Crypto support via CoinGecko (top 20 by market cap)
- Fundamental data via Financial Modeling Prep (P/E, revenue growth, margins)
- Earnings calendar with trendline enrichment
- Market regime filter (bull/bear/sideways detection) to gate entry signals
- Sector rotation signals
v4.2: Broker Integration + Human-Gated Live Trading
Goal: connect to a real broker (Alpaca) for paper and live trading with human oversight.
Deliverables:
- Alpaca API integration (paper trading mode first)
- Approval workflow: machine proposes → human approves → machine executes
- Real-time order execution at market open
- Trade journal with full audit trail
- Gradual auto-approve thresholds based on track record
v5.0: Autonomous Trading Within Bounds
Goal: machine handles routine decisions within pre-approved rules; human handles exceptions.
Deliverables:
- Rule-based auto-execution: “buy up to 3% of portfolio in ENTRY zone stocks with 2+ confirming signals”
- Portfolio-level risk management: max drawdown limits, correlation-aware position sizing
- Continuous strategy learning: compare signal predictions vs outcomes, adjust weights
- Anomaly detection: alert human when market behavior deviates from historical patterns
Original milestone framework
The original M1–M5 framework remains the architectural north star:
| Milestone |
Original Goal |
Current Status |
| M1. Foundation |
Domain models, controls, observability |
Partially shipped (schema in screener-api) |
| M2. Aggregation |
Multi-broker connectors, unified views |
Planned for v4.2 (Alpaca first) |
| M3. Advisor |
Portfolio intelligence, scenarios |
Shipped (screener + signal + navigator) |
| M4. Execution |
Approval workflows, trade routing |
Shipped (trade gate + paper trading) |
| M5. Learning |
Backtesting, strategy improvement |
Shipped (backtest engine + param sweep) |
PRDs for original milestones:
Progression to autonomous trading
TODAY Paper trading with daily pipeline, human reviews every trade
MONTH 1 Daily auto-pipeline via Vercel Cron, weekly human review
MONTH 2 Alpaca paper trading (real broker simulation)
MONTH 3 Alpaca live with small capital ($1-5K), human approves every trade
MONTH 6+ Gradual auto-approve within pre-set rules