Minara and Kronos debut AI trading tools

AI trading tools are getting more specific. That’s the real news here. Minara AI pushed a new Prediction Market Copilot for Bitcoin-linked prediction contracts on May 11, and Kronos is drawing attention from quants as an open model built just for candlestick data. The gap both are trying to fill is pretty obvious — traders are drowning in charts, order flow, and fast-moving markets, but most “AI trading” products still feel like black boxes or marketing wrappers. What changed is that these newer tools are being pitched less as autonomous money printers and more as working layers inside an actual trading workflow. ### What did Minara actually launch? Minara’s new product is called Prediction Market Copilot, and it’s built for Hyperliquid HIP-4 binary contracts tied to Bitcoin price outcomes. Instead of telling users to stare at a chart and decide manually, it outputs a simple YES, NO, or WAIT recommendation for each contract, then shows the estimated edge, confidence, expected value per dollar risked, and the technical reasons behind the call. Execution still runs through a connected Hyperliquid wallet with USDC, so the user remains the one pulling the trigger. (coinness.com) ### How does Minara make the call? Basically, it compares the market’s implied probability with its own AI-estimated probability. If a YES contract is priced at 0.58, that means the market is pricing the outcome at 58%. Minara then checks whether its model thinks the true odds are higher or lower, and the gap becomes the “edge percentage.” The system says it uses multi-timeframe candles plus indicators like RSI, MACD, Bollinger Bands, EMA, ATR, and support and resistance levels. That’s less “oracle” and more packaged technical-analysis stack with trade framing attached. (minara.ai) ### What is Kronos, then? Kronos is a different beast. It isn’t a retail copilot with buttons and prompts. It’s an open-source foundation model for financial candlesticks — what traders call K-lines — released as a family of decoder-only models with public weights on Hugging Face and code on GitHub. The core pitch is that finance data should be modeled like its own language, not treated as generic time series. So Kronos tokenizes OHLCV market data, then trains a transformer on that representation. (minara.ai) ### Why does the 12 billion number matter? Because scale is the whole claim. Kronos says it was pre-trained on more than 12 billion K-line records from 45 global exchanges. That matters less as a bragging-rights number and more because it hints at the model’s intended use: broad pattern recognition across assets, venues, and timeframes. The model card says Kronos is aimed at zero-shot forecasting, volatility prediction, and even synthetic data generation. In plain English, it wants to be infrastructure for quants and developers, not just a chatbot for traders. (github.com) ### So are these both “AI trading bots”? Not really — and that’s the interesting part. Minara’s own docs split the product into Copilot, Autopilot, and Strategy Studio. Copilot gives signals for the user to act on. Autopilot can execute within set limits. Strategy Studio is for writing and backtesting rules. That separation tells you the industry is maturing a bit. The new pitch is modular AI — one layer for signal generation, another for execution, another for systematic testing. (huggingface.co) ### Where do whale trackers fit in? They’re part of the same shift. Tools around Polymarket-style markets now offer whale alerts, anomaly scans, consensus scores, and smart-money tracking. Those products are less about predicting the future from scratch and more about surfacing information faster than a human dashboard can. Think of them as surveillance tools for market behavior — who’s buying size, where unusual flow is showing up, and whether sentiment is moving before price does. (github.com) ### What’s the catch? The catch is that workflow tools can still be wrong in very confident ways. Minara itself warns that expected-value numbers are estimates, not guarantees. Kronos may be powerful, but open models trained on historical candles still face the oldest market problem there is — regimes change, and patterns decay. A model that reads charts beautifully can still get run over by one macro headline or one liquidity shock. (polytrackers.com) ### Bottom line This story isn’t really “AI can trade now.” It’s that AI trading products are being rebuilt as interfaces for decision-making — scan the market, compress the signal, frame the risk, maybe execute if the user wants. That’s a smaller promise, but turns out it’s the more believable one. (minara.ai)