medes-et-bayse

An autonomous trading agent for Bayse Markets — Africa's largest prediction market platform. Uses PyFlue for agent orchestration with a Ralph loop pattern (Goal → Plan → Execute → Reflect → Repeat).

Architecture

User sends /goal in Telegram
       │
       ▼
  ┌─────────────────┐
  │   Ralph Loop     │ ← PyFlue agent entry point (bot/agent.py)
  │  Goal→Plan→     │
  │  Execute→Reflect │
  │  →Repeat         │
  └────────┬────────┘
           │
    ┌──────┼──────┐────────┐
    ▼      ▼      ▼        ▼
 Tavily  Bayse   Kelly   SQLite
 Search  Markets Criterion Memory
 (research) (scan) (signals) (log)

Strategies Implemented

• Kelly Criterion — Size positions optimally based on edge and bankroll
• Arbitrage Detection — Spot mispriced markets where Yes + No < 1 (implied probability gap)

Setup

pip install -r requirements.txt

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

cp .env.example .env

Environment Variables

Variable	Description
BAYSE_PUBLIC_KEY	Bayse Markets public key
BAYSE_SECRET_KEY	Bayse Markets secret key (HMAC-SHA256)
TAVILY_API_KEY	Tavily Search API key (for market research)
TELEGRAM_BOT_TOKEN	Telegram bot token from @BotFather
DATABASE_PATH	SQLite database path (default: /data/medes.db)
DRY_RUN	Set to true to simulate trades (default: true)
BANKROLL	Total capital allocated to the bot (USDC)
MAX_POSITION_FRACTION	Max fraction of bankroll per trade (default: 0.05)
MIN_EDGE	Minimum edge threshold before trading (default: 0.03)
CYCLE_INTERVAL_SECONDS	Seconds between Ralph loop cycles (default: 300)

Running the Agent

# Start the PyFlue agent with Telegram bot
python bot/agent.py

Telegram Commands

Command	Description
/goal	Start the autonomous Ralph loop
/stop	Stop the active loop
/status	Show agent status, recent trades, and notes
/trades	Show recent trade log

Ralph Loop Cycle

Each cycle:

1. Scan — Fetch open markets from Bayse API
2. Research — Use Tavily to gather news context for high-signal markets
3. Evaluate — Run Kelly Criterion + Arbitrage strategies
4. Execute — Place trades (or simulate in dry-run mode)
5. Log — Write everything to SQLite (trades, research, reasoning)
6. Report — Push cycle summary to Telegram chat

SQLite Memory Schema

Three tables at DATABASE_PATH:

• trades — Every trade with market_id, side, price, amount, edge, strategy, status
• market_research — Tavily research results tied to markets
• agent_notes — Agent reasoning and cycle summaries

Deployment (Render)

# Deploy as a Background Worker on Render
# render.yaml is pre-configured
render deploy

Requires a persistent disk mounted at /data for SQLite storage.

Project Structure

medes-et-bayse/
├── bot/
│   ├── agent.py          # PyFlue agent + Ralph loop + Telegram commands
│   ├── bayse_client.py   # Bayse Markets REST API client
│   ├── main.py           # Legacy entry point (strategies + polling)
│   ├── realtime_feed.py  # WebSocket/REST quote feed
│   ├── strategies/
│   │   ├── arbitrage.py  # Arbitrage detection
│   │   └── kelly.py      # Kelly Criterion sizing
│   ├── telegram_handler.py
│   └── tools/
│       ├── bayse.py      # Bayse API tool for PyFlue
│       ├── memory.py     # SQLite memory tool for PyFlue
│       └── search.py     # Tavily web search tool for PyFlue
├── medes_et_bayse/       # Package (client, auth, config, hermes loop)
├── PROMPT.md             # Agent goal definition
├── fix_plan.md           # Migration & cycle task plan
├── render.yaml           # Render deployment config
├── requirements.txt
└── .env.example