Agent Einstein forecasts prices with two independent machine-learning time-series models: Google TimesFM 2.5, a general-purpose time-series foundation model, and Kronos, a finance-native model trained on 12B+ K-lines from 45 exchanges. A consensus mode runs both in parallel and flags when they agree or diverge.
Every forecast is recorded at issuance — from automation, chat, the autonomous trading engine and the paid API alike — and scored against the actual close once its horizon elapses. This page publishes the resulting track record per model and horizon (MAPE, RMSE, direction hit rate), so the forecasts can be judged on evidence rather than claims.
The same models are available pay-per-use: the AI Chart Forecast service via x402, A2A, MCP and ACP, and an OpenAI-shape inference API at /api/einstein/v1/forecast, whose responses carry an EIP-191 attestation that lets a buyer prove offline what was predicted and when.