PROMETHEUS Trading Platform

An autonomous, self-improving AI trading system combining multi-source market intelligence, adaptive machine learning, and dual-broker live execution across equities, ETFs, and crypto.

Market Data (28 sources)
        │
        ▼
Real-World Data Orchestrator  ─────►  Intelligence Signals (persisted to DB)
        │
        ▼
Signal Voting Engine
  ├── HRM Transformer (fine-tuned, 85 epochs)
  ├── Local LLM (prometheus-trader via Ollama)
  ├── LangGraph 4-node decision graph
  ├── SB3 PPO Reinforcement Learning agent
  ├── LLaVA Vision (chart pattern analysis)
  ├── Technical Analysis (RSI/MACD/Bollinger/ATR)
  ├── Unusual Options Activity (sweep detection)
  ├── Insider Trading (SEC EDGAR filings)
  ├── Analyst Ratings (yfinance consensus)
  ├── StockTwits Sentiment (bull/bear ratio)
  ├── Dark Pool Activity (volume surge detection)
  ├── Options Chain (put/call ratio)
  ├── SEC Filings RAG (earnings transcripts)
  ├── Finviz Fundamentals (insider %, short float)
  ├── Fear & Greed Index (contrarian signal)
  ├── Put/Call Ratio (CBOE options flow)
  ├── RSS News Sentiment (10 live feeds)
  ├── Earnings Calendar (pre-event risk guard)
  ├── FOMC Calendar (macro event guard)
  └── Market Regime Detector (bull/bear/sideways/volatile)
        │
        ▼
Trade Decision → Dual-Broker Execution
  ├── Alpaca (equities, crypto, 24/5)
  └── Interactive Brokers (institutional execution)
        │
        ▼
Adaptive Learning Engine (5 continuous background loops)
  ├── Outcome Capture    — every 60s
  ├── Weight Updates     — every 5 min
  ├── Model Retrain      — every 1 hr
  ├── Insight Generation — every 15 min
  └── Risk Adaptation    — every 10 min
        │
        ▼
Parallel Shadow Trader ($100K virtual — learns alongside live trades)
        │
        ▼
prometheus_learning.db  (trade history, weights, signals, patterns)

PROMETHEUS learns from every trade automatically:

• AI voter weights update hourly based on which signals predicted correctly
• HRM checkpoint retrains on new labeled trade outcomes
• Ollama fine-tune triggers after 50+ winning trades — builds prometheus-trader model from actual wins
• Risk parameters (position sizing, stop loss) adapt to recent win/loss streaks
• Shadow trader runs $100K virtual capital in parallel — accelerates learning without risking real money

No human intervention required. The system compounds its own edge over time.

All SQLite with WAL mode. Do not delete — retraining from scratch takes hours.

These databases are NOT included in the repo (too large). See Migration Bundle below.

The prometheus-trader model is built automatically by core/ollama_finetuner.py after 50+ winning trades — no manual action needed.

• Python 3.11 — python.org
• Git — git-scm.com
• Ollama — ollama.com (install and leave running)
• Interactive Brokers Gateway — ibkr.com/gateway

git clone --recurse-submodules https://github.com/Awehbelekker/Prometheus.git
cd Prometheus

--recurse-submodules is required — pulls ThinkMesh reasoning engine

python -m venv .venv
.venv\Scripts\activate

NVIDIA GPU (CUDA):

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
python -c "import torch; print(torch.cuda.is_available())"  # → True

AMD GPU (DirectML):

pip install torch-directml

CPU only:

pip install torch torchvision torchaudio

pip install -r requirements.txt
pip install -e ThinkMesh/

copy .env.example .env
# Edit .env — fill in your API keys (see Environment Variables section below)

ollama pull llama3.1:8b
ollama pull llava:7b
ollama pull deepseek-r1:8b

The trained HRM model, learning database, and signal cache cannot be regenerated from code alone. Copy from an existing installation:

migration_bundle/
├── hrm_checkpoints/market_finetuned/   ← 270MB trained transformer (85 epochs)
├── prometheus_learning.db              ← trade history + learned weights
├── performance_metrics.db
├── trained_models/                     ← included in repo (1.7MB)
└── prometheus_real_hrm_signal_cache.npz  ← optional, saves 90min startup

Without hrm_checkpoints/market_finetuned/ the system falls back to LSTM — still functional but less accurate.

1. Open IB Gateway and log in
2. Configure: Settings → API → Enable socket port 4002 (live) or 4001 (paper)
3. Check "Allow connections from localhost only"

# Option A — double-click
run_prometheus.bat

# Option B — manual
.venv\Scripts\activate
python launch_ultimate_prometheus_LIVE_TRADING.py

# Option C — watchdog (auto-restarts on crash)
python prometheus_watchdog.py

Dashboard: http://localhost:8000

Copy .env.example to .env and fill in:

# ── Alpaca (required for stock/crypto trading) ────────────────────────────
ALPACA_LIVE_KEY=your_alpaca_live_key
ALPACA_LIVE_SECRET=your_alpaca_live_secret
ALPACA_LIVE_BASE_URL=https://api.alpaca.markets
ALPACA_PAPER_KEY=your_alpaca_paper_key
ALPACA_PAPER_SECRET=your_alpaca_paper_secret
ALPACA_PAPER_BASE_URL=https://paper-api.alpaca.markets

# Use live or paper:
ALPACA_API_KEY=your_alpaca_live_key
ALPACA_SECRET_KEY=your_alpaca_live_secret
ALPACA_BASE_URL=https://api.alpaca.markets   # or paper-api.alpaca.markets
ALPACA_PAPER_TRADING=false                   # true = paper only

# ── Interactive Brokers (optional, enhances execution) ────────────────────
IB_LIVE_ENABLED=true
IB_PORT=4002           # IB Gateway live port (NOT 7496 which is TWS)
IB_PAPER_PORT=4001
IB_HOST=127.0.0.1
IB_CLIENT_ID=1
IB_ACCOUNT=your_ib_account_number

# ── Local AI (required) ───────────────────────────────────────────────────
AI_PROVIDER=ollama
USE_LOCAL_AI=true
OLLAMA_BASE_URL=http://localhost:11434
USE_LLAVA_VISION=true

# ── Free data sources (recommended) ──────────────────────────────────────
FRED_API_KEY=your_fred_key        # free at fred.stlouisfed.org
REDDIT_CLIENT_ID=your_reddit_id   # free at reddit.com/prefs/apps
REDDIT_CLIENT_SECRET=your_reddit_secret
REDDIT_USER_AGENT=PROMETHEUS Trading Bot 1.0

# ── Optional / paid ───────────────────────────────────────────────────────
OPENAI_API_KEY=your_key           # fallback LLM
ANTHROPIC_API_KEY=your_key        # Claude vision
POLYGON_API_KEY=your_key          # market data
TWITTER_API_KEY=your_key          # social sentiment

Never commit your .env file. It is in .gitignore by default.

1. Download IB Gateway (lighter than TWS): ibkr.com/gateway
2. Log in with your IBKR credentials
3. Go to: Configure → Settings → API → Enable ActiveX and Socket Clients
4. Set socket port to 4002 (live) or 4001 (paper)
5. Check "Allow connections from localhost only"
6. Set IB_ACCOUNT= in .env to your account number (format: Uxxxxxxxx)

To copy a complete running installation over a local network:

# On new server — receive (run first):
net share PROMETHEUS_RECV=C:\Users\NewUser\Desktop\ /GRANT:Everyone,FULL

# On existing server — send:
robocopy "C:\Users\Judy\Desktop\PROMETHEUS-Trading-Platform" \\NEW-SERVER-IP\PROMETHEUS_RECV\PROMETHEUS-Trading-Platform /E /XD .venv_directml_test .venv-gpu311 __pycache__ .git /XF *.pyc

# Then copy the venv separately (or recreate fresh — recommended):
# Recreating fresh on new server is cleaner than copying venv

Or use the migration bundle approach — copy only what can't be cloned:

hrm_checkpoints/market_finetuned/
prometheus_learning.db
performance_metrics.db
.env

python prometheus_real_hrm_backtest.py
python prometheus_50_year_competitor_benchmark.py

• http://localhost:8000 — live P&L, positions, signal votes, system health
• admin_command_center.html — full admin panel

Trading involves substantial risk of loss. Past performance does not guarantee future results. Start with paper trading (ALPACA_PAPER_TRADING=true) to validate before committing real capital.