One execution core. One risk layer. Every venue. Ten strategy bots share a single engine and a venue-agnostic adapter stack — adding a market means writing one adapter, not rebuilding a bot. Twenty-one venues are mapped to that interface, each with its own repo.
Quick Start • Strategies • Managed Service • Venue Coverage • Engine • Safety • Contact
No Rust toolchain needed. Grab a prebuilt binary from the latest release — Linux (x86_64 / aarch64), macOS (Intel / Apple Silicon), Windows — verify the SHA256, and you're running in about a minute:
tar -xzf polymarket-toolkits-<tag>-<target>.tar.gz
cd polymarket-toolkits-<tag>-<target>
cp config.yaml.example config.yaml # credentials; config.json ships with the archive
./polymarket-toolkits run copy-trading # dry-run: enable_trading is false by defaultThe archive ships config.json (public settings) alongside config.yaml.example (credentials). Run ./polymarket-toolkits --help for the full command list, or launch with no subcommand for the interactive TUI.
Prefer to build it? cargo install --git https://github.com/tarpboi/Prediction-Markets-Trading-Bot-Toolkits
|
Open-source engine, your keys, your wallet. # 1. Clone the engine (the trading code lives here)
git clone https://github.com/tarpboi/Prediction-Markets-Trading-Bot-Toolkits
cd Prediction-Markets-Trading-Bot-Toolkits
# 2. Configure — credentials from the example
cp config.yaml.example config.yaml
# 3. Dry-run first (no real orders)
cargo run --release -- run copy-tradingEvery bot ships with |
Not sure which strategy fits your venue, capital size, or risk budget? Ask.
|
| 🎯 Strategies | ⚙️ Engine | 🧪 Dry-run |
|---|---|---|
| 10 | Rust | Every path |
Star and fork counts are live badges, not numbers typed into a README — they track the repo and can't drift. No fake testimonials, no cherry-picked P&L.
Ten strategies, one execution core: a venue adapter, an order executor with EIP-712 signing and a real dry-run path, hard spending limits, a risk guard, and position tracking. Pick the edge that fits your thesis — the infrastructure underneath is shared.
⚠️ Status: all ten run on Predict.fun in dry run against live books. No live order has been sent through any of them yet, and none has a track record. Dry-run P&L checks the plumbing and isn't a forecast. Stay in dry run until you've proven a strategy yourself with a small amount.
📦 Every bot runs on the shared engine and safety layer. The per-venue repos hold venue metadata — the trading code is here.
| # | Strategy | What it does | Key spec |
|---|---|---|---|
| 1 | 🎯 Copy Trading | Mirror one wallet's fills at your scale | Sizing and risk pipeline · TP/SL · circuit breaker |
| 2 | ⚡ BTC 5m / 15m / 1hr Arbitrage | Buy Up/Down rounds when the book lags the underlying | Taker · model value · held to round end, can lose |
| 3 | 💰 Cross-Market Arbitrage | Find gaps between linked markets on two venues | Predict.fun ↔ Polymarket · reports only, no live trading |
| 4 | 🎯 Directional Arbitrage | Rest Yes + No bids under $1, tilted toward the side with more edge | Limit-only · hedged only once both legs fill |
| 5 | 📈 Spread Farming | Capture the bid-ask spread, one round trip at a time | Post-only · hold limit · per-trade P&L |
| 6 | 🏆 Sports Execution | Type it. Filled. Done | Terminal tool · all-or-nothing · same limits |
| 7 | 🎯 Resolution Sniper | Buy near-certain outcomes about to end, hold to the payout | Certainty scan · a surprise costs dozens of wins |
| 8 | 📊 Orderbook Imbalance | The signal is the order book — no external feeds | Near-touch imbalance · target, stop and time limit |
| 9 | 💰 Market Making | Quote both sides, leaning away from your inventory | Post-only · inventory skew · expiry on every quote |
| 10 | ⚡ Whale Signal | Follow many wallets, or any large taker | Per-wallet risk buckets · venue trade feed |
How the flagship edges actually work (click to expand)
🎯 Copy Trading — Point the bot at a wallet with a verifiable record. It mirrors the wallet's fills at your chosen scale, with exposure caps and a circuit breaker that halts on abnormal bursts.
💰 Cross-Market Arbitrage — Predict.fun publishes which Polymarket market each of its markets is linked to, so the engine pairs them on that link rather than on matching titles, and reports a gap only when it beats fees on both venues. It doesn't trade: a live version has to send both legs, and one filled leg without the other is a naked position. Check that both venues resolve a question the same way before trusting a link.
🎯 Directional Arbitrage — Rest bids on Yes and No that total under $1, then tilt extra size onto the side the book favours. The floor only applies once both legs have filled: until then it's an ordinary directional position. On a venue with one shared book, like Predict.fun, this has to be built from resting orders, because taking both asks always costs more than $1.
🎯 Resolution Sniper — Scan markets about to end for an outcome at 95¢ or more, and hold to the payout. Each win is small and each loss costs nearly the whole stake, so buying at 97¢ only pays if the outcome is right more than 97% of the time. The market's price is the only evidence: it doesn't check the real-world result.
📊 Orderbook Imbalance — No external feeds, no oracle: the signal is the book. Near-touch depth skew becomes a short-term directional read, polled every few seconds. Displayed size is cheap to fake, and every entry pays the spread.
💬 Want a strategy explained for your venue or capital size? → t.me/tarpboi
⏸️ The hosted service is not currently open. It ran as a paper-trading beta — simulated funds, no real custody — and is offline while the infrastructure moves. It never handled real money, and live trading was always gated behind custody, security audit and licensing. Sign-ups and pricing are closed for now, so the way to use this project today is to run it yourself with your own keys (see Run the bots yourself).
If you want to talk about a managed setup when it reopens — or about running the engine at size yourself — that conversation happens on Telegram:
The engine is venue-agnostic by design: any platform exposing an order book or position feed plugs in through a single adapter, so adding a market means writing one adapter rather than rebuilding a bot.
Twenty-one venues are mapped to that interface, each with a repo of its own. The per-venue repos hold venue metadata, so clone this repo for the trading code.
| Venue | Type | Strategies |
|---|---|---|
| Predict.fun | Decentralized (BNB Chain / USDT) | All ten — dry-run tested on testnet and mainnet books; live orders not yet tested (Cross-Market reports gaps only) |
| Polymarket | Decentralized (Polygon / USDC) | Copy Trading — order path still signs the V1 format and needs updating for CLOB V2 |
Predict.fun settles in USDT on BNB Chain and uses the same CTF-exchange contract design as Polymarket, with short crypto rounds. Signatures and limit-order amounts match Predict's official SDK byte for byte, and every strategy has run in dry run against the testnet API, and all but Copy Trading against mainnet books as well. Live order submission hasn't been tested yet, so stay in dry run until you've proven it with a small amount.
# Build from source: this isn't in a release yet. Defaults to testnet and dry run.
cargo run --release -- --venue predict-fun check-venue
# Any of: copy-trading btc-arb cross-arb directional-arb spread-farming sports
# resolution-sniper orderbook-imbalance market-making whale-signal
cargo run --release -- --venue predict-fun run market-makingCopy Trading needs a wallet in predict_fun.wallets_to_track in config.json. Tune each strategy and its hard limits under strategies there. The filters allowlist applies to these bots too, and the shipped one (Politics, Sports, Crypto) leaves out most of Predict's markets, so review it first. BTC rounds settle on Binance's top of book: where Binance isn't reachable, that bot uses a Coinbase price and stays on paper.
Mainnet needs "network": "mainnet" in config.json and an API key from Predict. Before trading live, set approvals once with the SDK's setApprovals, use a dedicated wallet, and read Predict's Terms of Service: they bar several jurisdictions, including the US and UK, and signing in accepts them. The bots don't redeem winnings, so redeem them on Predict. In October 2026 Predict reported about 400,000 USDT in unauthorized wallet transfers; the venue repo has details.
| Venue | Type |
|---|---|
| Kalshi | CFTC-regulated (US) |
| PredictIt | Academic / US politics |
| Robinhood Predictions | Brokerage-integrated |
| Crypto.com Predictions | Crypto-integrated |
| OG.com | Social / multi-outcome |
| DraftKings Predictions | Sports |
| FanDuel Predicts | Sports |
| Fanatics Markets | Sports / entertainment |
| Interactive Brokers ForecastTrader | Financial events |
| Venue | Chain / Type |
|---|---|
| Limitless | On-chain order book |
| Drift BET | Solana |
| Azuro | Decentralized protocol |
| Augur | Ethereum |
| Myriad Markets | Crypto |
| Hedgehog Markets | Solana / social |
| Zeitgeist | Polkadot |
| Projection Finance | Volatility / sims |
| Better Fan | Sports / esports |
| Manifold Markets | Play-money · consensus signal |
Want a venue prioritized? Adapter work is demand-driven — if you trade a platform you'd like wired up next, reach out on Telegram and it can move up the queue.
Rust, async on Tokio, one execution core behind every strategy and venue. The adapter stack means a new market is one adapter — not a new bot.
| Event processing | < 1ms per event |
| Order execution | < 100ms end-to-end |
| Position polling | ~200ms per wallet |
| Memory | ~50MB baseline |
| CPU | < 5% on modern hardware |
| Concurrency | Semaphore-based rate limiting (default: 25 req / 10s) |
| Circuit Breaker | Auto-halts after N consecutive large trades inside a configurable window |
| Depth Guard | Validates orderbook liquidity before every order |
| Dry Run | Full execution path runs without placing real orders |
| Trade Floor | Minimum size enforcement against negative-EV micro-trades |
| Spending Limits | Hard caps per order, per market and in total, plus a loss limit that halts new entries. Every strategy order passes through them |
| Existing Positions | What a wallet already holds counts toward the limits, and no strategy will sell it |
The circuit breaker fires when consecutive large trades exceed the configured threshold, or when orderbook depth falls below the minimum. Once tripped, execution is blocked for the cooldown duration. Trip state and cooldown are logged and visible in the TUI.
Recommendations:
| Stage | Action |
|---|---|
| Initial setup | Run with enable_trading: false for a full session |
| First real trades | Keep copy_percentage at 5–10% until you trust the signal |
| Ongoing | Watch circuit breaker trips — they surface execution anomalies |
| Production | Dedicated wallet with only the capital you intend to deploy |
Built and maintained actively. Whether you want to run the bots, request a new venue adapter, or just talk Polymarket tooling and algorithmic strategies — reach out.
| Platform | Link |
|---|---|
| Telegram | t.me/tarpboi |
| Discussions | GitHub Discussions |
Response time is typically within a few hours. Open to questions, feedback, venue requests, and serious collaborations.
Trading prediction markets involves real financial risk. This software is provided as-is, without warranty or guarantee of any outcome. It is not financial advice. Always test with
enable_trading: falsebefore deploying real capital. Check the source for the current state of each strategy before running it with real capital. The managed / copy-trading service is currently closed — it ran in paper mode (simulated funds), never custodied real money, and any future live-trading rollout will follow proper custody, audit, and licensing. Ensure compliance with each venue's terms of service and applicable regulations in your jurisdiction.
Built for the Prediction Markets including Predict.fun, Polymarket, Kalshi, Limitless etc
