polymarket-arbitrage: what it does, what the screenshots mean, where it stops

What it does

In real mode it uses Polymarket's Gamma API for market discovery and the CLOB API for order books, and pulls Kalshi public market data with no authentication. A FastAPI dashboard at localhost:8000 shows opportunities, markets monitored, order books, uptime and PnL.

The package includes a risk manager with position limits, a daily loss limit and a kill switch, a portfolio tracker, a backtest utility and unit tests.

• config.yaml controls trading_mode (dry_run or live), data_mode (real or simulation), min_edge (default 1%), min_match_similarity (0.6)
• Risk defaults: $200 per market, $5,000 global exposure, $500 daily loss
• Market-making trigger: spread of 5 cents or more

Strategy

Cross-platform: if the same prediction is cheaper on one venue, buy there and sell on the other; the README example is 52 cents on Polymarket against 58 cents on Kalshi. Bundle: buy both sides when asks sum below $1.00, sell both when bids sum above it.

Fee accounting is in config, with Polymarket maker and taker fees set to 0 basis points and gas at $0.001 per order. The README states that real markets are efficient and opportunities are rare.

Setup and requirements

Python 3.10 or newer, pip install -r requirements.txt, then python run_with_dashboard.py. Live trading needs POLYMARKET_API_KEY and POLYMARKET_PRIVATE_KEY as environment variables. Kalshi reads are unauthenticated; Kalshi trading requires a US account with KYC, which the README notes.

Activity and maintenance

274 stars and 122 forks as of 2026-09-08. Created 2025-12-07, last push 2025-12-09, 4 open issues. Nine months without a commit: unmaintained. The README's MIT badge is not backed by a LICENSE file.

Where it falls short

Cross-platform selling requires holding the position on the venue you sell on; the bot's table describes buy-on-one, sell-on-other without addressing that. Text similarity at 0.6 will pair markets that resolve differently.

The only performance evidence is from simulation. Polymarket's fee model and CLOB have changed since December 2025 and the code has not.

Who should use it

Readers. It is a clean, well-organized reference for how the three arb families are detected and risk-limited. Fork it for the structure, then rewrite the venue clients.

Not for live capital without a rewrite.