Forecast predicts what is coming — sales next quarter, inventory you will need, cash flow, customer volume. Upload your past numbers, pick what you want to predict, and get a clear forecast with a confidence range. No spreadsheet formulas. No data science team.
Configure once, run continuously. Every capability below is a production-grade surface, not a demo.
Google's TimesFM 2.5 handles the forecasting. No ARIMA tuning, no LSTM wrangling — it generalises across series with enough history to be useful out of the box.
Data sources, cleaning, seasonality, covariates, anomaly detection, scenario modelling. Each node has a visible contract; pipelines stay auditable.
Every forecast ships with a band. Plan around the interval, not a single line — overstock drops, stock-outs drop, decisions get honest.
Inject holiday calendars, promo schedules, weather, or any exogenous series. The model learns the relationship; the pipeline keeps running after launch.
An online home-goods brand uploads 18 months of sales and adds holiday and promotion dates. Forecast predicts the next 12 weeks of demand for each product — with a confidence range, not a guess.
Every Forecast module shares the same orchestration layer, the same audit trail, and the same integration fabric — so a win in one surface lands everywhere.
Google's TimesFM 2.5 handles the forecasting. No ARIMA tuning, no LSTM wrangling — it generalises across series with enough history to be useful out of the box.
Data sources, cleaning, seasonality, covariates, anomaly detection, scenario modelling. Each node has a visible contract; pipelines stay auditable.
Every forecast ships with a band. Plan around the interval, not a single line — overstock drops, stock-outs drop, decisions get honest.
Inject holiday calendars, promo schedules, weather, or any exogenous series. The model learns the relationship; the pipeline keeps running after launch.
Data source → preprocessing → forecast → chart. The node metaphor is shared with Painter; the domain is time series, not media. Same foundations, different node library.
A single chart that's actually honest: solid history, dashed projection, shaded confidence. Zoom, pan, diff two runs side by side.
Every pipeline run is a record. Fork it, change the preprocessing, rerun — then compare accuracy over holdout windows.
CLI, SDKs and a single HTTP API. Every product on the platform exposes the same surface — same auth, same idempotency, same observability.
# Run a scheduled forecast pipeline
mariete forecast run \
--pipeline weekly-revenue \
--covariates holidays,promos \
--horizon 90dWhat comes in, what goes out, what it runs on. Nothing hidden in a sales deck.
Add competitors by URL. Collectors, synthesis, daily briefings, weekly reports, triggered alerts.
Explore Radar
Source material in, hundreds of cognitive agents out. Run parallel simulations; watch opinions shift.
Explore SimulateStep-by-step recipes that wrap Forecast around a specific outcome. Same agent, plugged into a real workflow your team already cares about.
Upload your sales history and add your holiday and promotion dates. Forecast predicts the next twelve weeks of demand for every product you sell — with a confidence range, not a guess.
Forecast learns what your normal cancellation and revenue pattern looks like — then pings you the moment the numbers start drifting.
Forecast pulls your invoicing data, cleans out the seasonal bumps, and writes a rolling six-month revenue forecast by service line.
Forecast predicts how many customers each location will see, hour by hour — factoring in weather, local events, and time of year. That drives the schedule.
No credit card. No sales call. Run one workflow end-to-end and decide.