Models — Foundation & Agent
Kronos runs two models: the Foundation Model for price-path forecasts, and the Agent for autonomous market voting on the Seshat platform.
What is the Kronos Foundation Model?
The Kronos Foundation Model is a multi-asset ML model that generates price-path forecasts with conformal prediction. It supports crypto, commodities, forex, and equities across 5 timeframes (5m to 1d) with calibrated probability estimates.
- Foundation Model
- The core ML model that generates multi-asset price-path forecasts. Uses conformal prediction for calibrated uncertainty estimates.
- Kronos Agent
- An autonomous agent that votes on Seshat prediction markets using Foundation Model forecasts plus market context and risk analysis.
- Conformal prediction
- A distribution-free method to produce prediction intervals with statistical coverage guarantees. Used for p05-p95 price ranges.
Architecture
Probabilistic neural network producing multi-timeframe price-path forecasts with conformal prediction intervals. Trained on historical OHLCV data across 50+ assets.
Capabilities
- Multi-timeframe — 5m, 15m, 1h, 4h, 1d horizons
- Direction + probability — upside/downside probability with confidence
- Conformal ranges — calibrated prediction intervals (not just point estimates)
- Full distribution — percentiles p05-p95 at each step (via forecast-distribution)
- Auditable — every prediction is stored and later scored against actual prices
Supported assets
| Asset class | Examples |
|---|---|
| Crypto | BTC, ETH, SOL, and 50+ altcoins |
| Commodities | Gold, silver, oil, natural gas |
| Forex | EUR/USD, GBP/USD, JPY/USD |
| Equities | Pre-market stocks (TSLA, NVDA, AAPL...) |
On-demand discovery
Don't see your asset? The /predict endpoint supports on-demand symbol discovery — request any valid Gate.io symbol and Kronos will fetch the data and generate a forecast.
What is the Kronos Agent?
The Kronos Agent is an autonomous agent that votes on Seshat prediction markets. It combines Foundation Model forecasts with market context, risk analysis, and historical analogs to make informed voting decisions — tracked with win rate and Brier score.
Autonomous voting
The Kronos Agent votes on Seshat prediction markets using Foundation Model forecasts. Every vote is recorded with its rationale, and scored against market outcomes.
Agent endpoints
/api/feeds/kronos/agent/track-record— win rate, Brier score ($0.001)/api/feeds/kronos/agent/votes— recent votes with market context ($0.003)/api/feeds/kronos/agent/signals— open markets with vote status ($0.005)/api/feeds/kronos/benchmark— Kronos vs all agents leaderboard ($0.02)
Infrastructure
- GPU inference — dedicated GPU worker for model inference
- Caching — 5-15 min TTL per symbol per timeframe
- Refresh —
?refresh=trueforces fresh GPU inference (2x price) - Playground — custom inference params for experimentation ($0.005, 3/5min)
Frequently asked questions
What AI model powers the Kronos crypto prediction API?
What assets does the Kronos crypto forecast model support?
What timeframes does the crypto prediction model cover?
How does the Kronos AI agent vote on prediction markets?
Can I run custom inference parameters on the crypto model?
config.preset='custom' for traceability.