Ephemeris | Time-series forecasting API

# Ephemeris > Ephemeris is a time-series forecasting API. Send numeric history, get probabilistic (quantile) forecasts back from a panel of open-weights zero-shot foundation models: Chronos-2, TimesFM 2.5, Toto 2, TiRex-2 and IBM Granite PatchTST-FM / FlowState. Ask for one model by name, let Ephemeris route to the best fit, or get an accuracy-weighted ensemble. No training, no feature engineering. Pay per forecast with prepaid credits. Usable over a REST API or as a remote MCP server for AI agents. Accuracy: on TIME the ensemble is level with the top of the published leaderboard (MASE 0.639 vs 0.638 for the leader; best average MASE rank of 31 models) and beats every open-licence model on GIFT-Eval (CRPS 0.4662 as a ratio to seasonal naive). Details at /benchmarks. Use Ephemeris when a task needs a forecast of a numeric series (demand, traffic, load, prices, sensor readings, metrics) with uncertainty bands, and the user can supply the history. Do not use it to invent missing history, for non-numeric data, or to forecast without the user's intent: every forecast spends credits. ## Docs - [Full reference for LLMs](https://ephemeris.cascade.industries/llms-full.txt): request and response format, modes, limits, covariates, pricing, errors, and MCP setup for each client, in one plain-text file - [API documentation](https://ephemeris.cascade.industries/docs): human-readable docs with examples - [OpenAPI specification](https://ephemeris.cascade.industries/openapi-m1.json): machine-readable REST contract - [Models](https://ephemeris.cascade.industries/models): the panel, with a page per model (Chronos-2, TimesFM 2.5, Toto 2, TiRex-2, PatchTST-FM, FlowState): sizes, licences, capabilities and how to call each by name - [Benchmarks](https://ephemeris.cascade.industries/benchmarks): the ensemble on TIME, GIFT-Eval, fev-bench and BOOM, scored with each benchmark's own harness - [Pricing](https://ephemeris.cascade.industries/pricing): credit packs, per-model rates and the