AI trades prediction markets.
Three AI models estimate probabilities. Kelly criterion sizes every bet. Settlement arbitrage catches mispricings in real-time. 12 autonomous loops run 24/7 on Kalshi — you set the limits, the AI handles the rest.
CFTC-regulated exchange · Real money · Risk controls built in · Not suitable for all investors
Want stock trading signals instead? See Tradewink Signals →
Why Tradewink Predictions
Edges that compound
Most prediction market traders use gut feel or a single model. This system stacks multiple quantitative edges — ensemble AI, optimal sizing, event detection, and flow analysis.
3 AI models, 1 consensus
Claude Haiku, GPT-5 Mini, and GPT-5 estimate independently via OpenRouter. Geometric mean produces calibrated probabilities designed to outperform any single model in aggregate — though accuracy is not guaranteed.
Math-optimal bet sizing
Kelly criterion with fractional scaling, bankroll limits, and confidence gating. Sizes every bet according to the mathematical formula — not gut feel. Optimal sizing does not guarantee profitable outcomes.
Faster than the market
Settlement arbitrage detects Fed decisions, CPI releases, and earnings in seconds. Attempts to act in the 5-10 minute repricing window — execution depends on market conditions.
Sees what others miss
VPIN analysis detects informed trading flow. Identifies order imbalances that may signal institutional positioning — not guaranteed to predict direction.
Self-correcting accuracy
Brier decomposition + adaptive Platt scaling. The system measures its own calibration and recalibrates based on actual outcomes. Accuracy improves in stable regimes but can degrade in rapidly changing markets.
Risk-managed by default
Risk parity across categories, max bet limits, and bankroll protection. Designed so no single market wipes the account — though cumulative losses across correlated markets remain possible.
Market Categories
Every market the AI can trade
From Fed rate decisions to weather records. The AI researches each category with specialized data sources and calibrated probability models.
Economics
Fed rate decisions, CPI, jobs reports, GDP growth. AI ingests macro data in real-time and estimates probabilities across possible outcomes.
Politics
Elections, legislation, Supreme Court rulings. Multi-model AI deliberation cross-references polls, news, and historical patterns.
Sports
Game outcomes, player milestones, championship futures. AI cross-references odds, injury reports, and historical matchup data for calibrated probabilities.
Climate
Temperature records, hurricane landfalls, precipitation thresholds. Open-Meteo weather forecasts feed directly into probability estimates.
Crypto
Bitcoin price thresholds, ETF approvals, regulatory actions. AI tracks on-chain metrics and sentiment across multiple sources.
Financials
S&P 500 levels, earnings surprises, IPO outcomes. The same data pipeline that powers stock trading, adapted for binary markets.
Culture
Awards shows, viral moments, media milestones. AI estimates probabilities where prediction markets may be thin or imprecisely priced.
Tech & Science
Product launches, scientific milestones, regulatory approvals. AI tracks news feeds and research timelines to estimate outcomes.
6 Strategies
How the AI finds and exploits edge
Not one trick — six complementary strategies running simultaneously. Each one targets a different source of alpha in prediction markets.
Ensemble Forecasting
Claude Haiku, GPT-5 Mini, and GPT-5 estimate independently via OpenRouter. Geometric mean of odds produces calibrated, externally Bayesian forecasts.
Kelly Criterion Sizing
Optimal bet sizing using fractional Kelly. Automatically adjusts for YES and NO sides, bankroll fraction, and confidence calibration.
Smart Money Detection
VPIN (Volume-Synchronized Probability of Informed Trading) detects institutional flow. Identifies order imbalances that may signal informed trading activity.
Settlement Arbitrage
5-10 minute repricing window after Fed decisions, jobs reports, and earnings. AI detects events and attempts to trade the repricing window.
Cross-Platform Arb
Cross-platform price divergence detection. When the same event is priced differently across exchanges, the AI attempts to trade the spread.
Market Making
Avellaneda-Stoikov adapted for binary markets. Provides liquidity with inventory-aware bid/ask quotes. Designed to collect the bid/ask spread.
The Pipeline
Data in. Probabilities out. Bets placed.
Multiple AI models and data feeds converge into a single estimation engine. When edge is found, bets are sized and executed automatically.
Data Sources
Kalshi
Live market prices
Claude Haiku
Research estimator
GPT-5 Mini
Quant reasoning
Open-Meteo
Weather forecasts
News Feeds
Real-time headlines
Prediction Engine
3-Model Ensemble
Execution
Kelly Sizing
Optimal bet size
Maker Orders
GTC for low fees
Settlement Arb
Event-driven
Calibration
Self-improving
How It Works
Set your limits. AI does the rest.
Configure your bankroll, risk tolerance, and market preferences once. 12 autonomous loops handle scanning, estimating, sizing, and executing around the clock.
AI scans every market
Hundreds of prediction markets monitored continuously. Each one gets full AI research — news, data, forecasts, and multi-model probability estimates.
Edge detected, bet sized
When the AI estimate diverges from market price by enough, Kelly criterion calculates the optimal bet. Risk parity ensures diversification across categories.
Executed automatically
GTC maker orders for lower fees. Settlement arb trades the repricing window automatically. Smart money signals trigger immediate action. All fully autonomous.
Getting Started
Connect your Kalshi account
Add your Kalshi API key and private key. The system connects directly to the exchange for real-time data and order execution.
Pricing
Start free. Scale as you grow.
Free plan gives you 5 AI probability estimates a day — no credit card needed. Paid plans unlock multi-model ensemble estimation and unlimited bets. Starter includes a 7-day trial; Pro and Elite start immediately.
Free
Try autonomous prediction trading with AI probability estimates.
Starter
Upgraded AI models for better probability estimates and auto-execution.
Pro
Multi-model deliberation + REST API for serious prediction traders.
Elite
Flagship AI models for highest-conviction prediction trading.
Common questions
Currently Kalshi, the only CFTC-regulated prediction market exchange in the US. The system trades binary event contracts across economics, politics, sports, climate, crypto, financials, culture, and tech & science categories.
Three AI models (Claude Haiku, GPT-5 Mini, GPT-5) independently research and estimate each market via OpenRouter. Their outputs are combined using the geometric mean of odds — a method that satisfies external Bayesianity and outperforms simple averaging.
A mathematically optimal bet sizing formula that maximizes long-term bankroll growth. The system uses fractional Kelly (typically 0.25x) to be conservative, and adjusts for both YES and NO side bets.
When a high-impact event resolves (e.g., Fed rate decision), prediction markets take 5-10 minutes to fully reprice. The AI detects events in real-time and trades the mispricing window before the market catches up.
VPIN (Volume-Synchronized Probability of Informed Trading) measures order flow imbalance. When one side dominates trading volume, it signals informed traders are moving. The system detects and follows these flows.
Yes. 12 concurrent agent loops run 24/7 — scanning markets, estimating probabilities, sizing bets, executing trades, tracking calibration, and scanning for arbitrage. You set your risk limits and the AI handles everything.
Prediction markets involve real money and real risk. Past performance does not guarantee future results. The system has risk controls (Kelly sizing, bankroll limits, risk parity), but losses are possible — including significant losses. Only trade with money you can afford to lose. Prediction market trading is not suitable for everyone.
You need a Kalshi account to use Predictions. Kalshi is available to US residents who meet their eligibility requirements (age 18+, identity verification). Kalshi charges trading fees on each contract — see kalshi.com for current fee schedules. Tradewink does not control Kalshi's fees, eligibility rules, account minimums, or market availability.
Yes. Paper trading mode lets you run the full AI pipeline without placing real money bets. It's recommended for new users before enabling live trading. You can toggle paper mode in your Predictions settings at any time.
AI models have real limitations. Black swan events, unprecedented geopolitical shocks, and sudden market microstructure changes can cause significant losses. The ensemble approach reduces single-model error but does not eliminate it. Models are calibrated on historical data and may underperform during novel or rapidly changing conditions. The system cannot guarantee any outcome.
Tradewink Autopilot
Want AI to trade stocks for you?
Predictions trades event contracts on Kalshi. Core trades stocks and options on your brokerage account — with 30+ brokers, autonomous day trading, and self-improving strategies.
Tradewink Signals
Prefer trade ideas over full autopilot?
Signals delivers AI-grade stock research to Discord or email — 16 signal types with full explanations. You make the call.
Let the AI play the odds
Ensemble forecasting, optimal sizing, and real-time arbitrage — running autonomously on the only CFTC-regulated prediction market exchange.
Tradewink is not a registered investment adviser, broker-dealer, or financial planner. All data, signals, and analytics on this page are for informational purposes only and do not constitute investment advice, financial advice, or a recommendation to buy or sell any security.
Past performance does not guarantee future results. Trading involves substantial risk of loss, including the possibility of losing more than your initial investment. You are solely responsible for your own trading decisions.
Hypothetical or backtested performance results have inherent limitations. Unlike actual trading records, simulated results do not represent real trading and may not account for the impact of market liquidity, slippage, or all transaction costs. No representation is made that any account will or is likely to achieve profits or losses similar to those shown.