Polyclaw Agent — Open-Source Arbitrage Bot for Polymarket

Overview

Polyclaw Agent is a free, open-source arbitrage bot for Polymarket, released under the MIT license. It does one thing: find and execute arbitrage opportunities on Polymarket’s Central Limit Order Book. No sentiment analysis, no momentum trading, no multi-platform support. Just arbitrage.

The bot was created by a solo developer in late 2025 and open-sourced after several months of personal use. The codebase is intentionally small — roughly 2,500 lines of Python — making it readable, auditable, and straightforward to modify. If you want to understand how a prediction market arbitrage bot works at the code level, Polyclaw is one of the most accessible starting points available.

Polyclaw supports two arbitrage strategies: Yes/No spread detection on binary markets (buying both sides when the combined cost is below $1.00) and cross-market arbitrage on multi-outcome markets (buying all outcomes when the implied probability sum is below 100%). It does not support cross-platform arbitrage (Polymarket to Kalshi), sentiment-driven trading, or any non-arbitrage strategy.

This narrow focus is a feature, not a limitation. Polyclaw is built for developers who want a clean, understandable arbitrage engine they can run as-is or use as a foundation for custom extensions. It’s not trying to be a full trading platform.

For a deeper understanding of how arbitrage works on Polymarket, our arbitrage bot case study breaks down a real six-month track record.

Key Features

Supported Strategies

Yes/No Spread Detection

Polyclaw continuously monitors Polymarket’s binary markets via the CLOB WebSocket API. For each market, it tracks the best ask price on both the Yes and No contracts. When the combined cost of buying one Yes share and one No share drops below $1.00 minus a configurable minimum spread, the bot executes both legs.

The execution logic handles partial fills gracefully. If the Yes leg fills but the No leg’s price has moved adversely, the bot either:

• Waits up to a configurable timeout for the No price to return to an acceptable level, or
• Exits the filled Yes position at market price if the timeout expires

The minimum spread, maximum position size, and fill-wait timeout are all configurable in the YAML config file. Default settings require a minimum spread of 0.8% and cap position sizes at 500 shares per trade.

Multi-Outcome Cross-Market Arbitrage

For multi-outcome markets (e.g., “Who will win X?” with multiple candidates), Polyclaw tracks the Yes ask price for every outcome. When the sum of all Yes prices drops below $1.00 minus the minimum spread, the bot buys one share of every outcome — guaranteeing a profit at resolution because exactly one outcome will pay $1.00.

This strategy requires more capital per trade (buying N outcomes instead of 2) and is subject to the liquidity of the thinnest outcome. Polyclaw sizes the trade to the lowest available depth across all outcomes, which often limits positions to small sizes on less popular candidates.

The bot skips multi-outcome opportunities where any single outcome has fewer than the configurable minimum depth (default: 50 shares).

Pricing Tiers

There are no paid tiers. The MIT license permits commercial use, modification, and redistribution. You can fork the repository, modify the code, and sell your own version if you want — the license imposes no restrictions beyond copyright attribution.

For context on how free agents compare to paid alternatives, see our prediction market bot pricing guide.

Performance

Polyclaw’s developer published a personal performance summary from their own usage (August 2025 through January 2026, six months). These are self-reported figures from a single operator and should not be treated as guaranteed or typical results.

Important caveats: These are a single operator’s results under specific market conditions during a specific time period. Arbitrage profitability has been declining as competition increases — spreads available in August 2025 were wider than those available in March 2026. Your results will depend on your capital, configuration, infrastructure, and the competitive landscape at the time you run the bot. This is not a guarantee of performance.

Pros and Cons

Pros:

• Completely free under a permissive MIT license
• Small, readable codebase (~2,500 lines) — easy to understand, audit, and modify
• Minimal dependencies keep the attack surface small
• Focused design — does arbitrage well rather than doing many things poorly
• Structured JSON logging makes it easy to analyze performance and debug issues
• Good starting point for developers who want to learn how arbitrage bots work
• MIT license allows commercial use and modification without restrictions

Cons:

• Polymarket only — no Kalshi support, no cross-platform arbitrage
• Arbitrage only — no sentiment, momentum, market making, or other strategies
• Self-hosted only — requires you to manage your own infrastructure (server, uptime, monitoring)
• No web dashboard — configuration is via YAML files, monitoring is via logs
• Community support only — no dedicated support team, response times vary
• Single developer project — bus factor risk; sustainability depends on maintainer activity
• No backtesting engine — you’ll need to build your own or use an external tool

How to Get Started

1. Clone the repository from GitHub. The README includes installation instructions.
2. Install Python 3.11+ and run pip install -r requirements.txt. The dependency list is intentionally short.
3. Copy config.example.yaml to config.yaml and fill in your Polymarket API credentials (API key, secret, passphrase). The config file documents every parameter with comments.
4. Run the opportunity scanner first — python polyclaw.py --scan-only runs the detection logic without executing trades. Let it run for a few hours and review the log output to see what opportunities the bot would have taken.
5. Start paper trading — python polyclaw.py --paper uses live market data but logs trades without executing them. Run for at least a week.
6. Go live — python polyclaw.py runs the full detection and execution pipeline. Start with conservative settings (high minimum spread, small max position size) and tighten gradually as you gain confidence.

For the full picture on setting up and securing a prediction market trading bot, see our guides on building a prediction market agent and agent betting security.

Similar Agents

• PredictEngine Pro — Commercial multi-strategy bot with Polymarket and Kalshi support. Hosted service, easier setup, but costs $49-$299/mo and you can’t inspect the strategy code.
• OctoBot Prediction Markets — Open-source modular bot with arbitrage, market making, and sentiment modules. More feature-rich but larger codebase and more complex to configure.

Browse all available agents in the Agent Marketplace. For a broader view of what’s available, see our best prediction market bots guide.

Listing last updated March 2026. Agent ratings are based on aggregated user reviews and editorial assessment. AgentBets does not receive compensation from listed agents. See our bot verification guide for how we evaluate agents.