Models
A different kind of forecast model.
Metis trains time-series foundation models, pretrained on a vast corpus of real-world series, so they forecast yours without fitting a new model for each one.
Two models.
A general foundation model, and a power specialist built on it. Metis-Power-Preview takes Metis-Base and trains it deep on energy, ISO markets, gas, oil, and the events that move them. Both ship as families from 10M to 1B parameters. Open a row for the full card.
Not a faster XGBoost.
A foundation model isn't one more gradient boost or another model per series. It's a different starting point, the same shift that reshaped language and vision, arriving for time series, and it answers with a distribution, not a single guess.
XGBoost & classical ML
- XGBoost, classical ML, and ARIMA, one model per series
- Features engineered by hand for each target
- Retrained constantly as the series drifts
- Point forecasts, weakest on the rare regimes that cost the most
A foundation model
- One pretrained model that forecasts a new series on contact
- Context, covariates and events, read directly rather than hand-built
- A calibrated distribution over outcomes, native to the model
- A cross-series prior that has seen regimes yours hasn't
The models are one half.
On their own, the models forecast. What we sell is the platform that turns a forecast into the call you make: Metis-TSFM and Metis-Agent, together.
The forecast engine
Metis-Base and Metis-Power read the system and return calibrated forecasts of what happens next.
The decision layer
Scores your options against your objective and risk limits, returns the call, and plugs into your stack over API and MCP.
Together, that's Metis, the decision layer for energy.
The forecast and the decision in one loop, graded on the baselines you already trust. See it on the desk →
Open source.
We build and open-source tooling for the time-series ecosystem, the parts of the stack the whole field benefits from. Here's what's public.
| Project | What it is | Source |
|---|---|---|
| ForecastOps fops | Local-first forecast observability and evaluation. Capture, validate, compute metrics into DuckDB, explore with fops ui. | GitHub → |
Put a foundation model on your forecasts.
Bring a series you already forecast and the baseline you trust, and we'll show what Metis does with it.