pretickGeneral time-series models
Chronos-2, TimesFM 2.5 and TiRex-2 provide different forecasting approaches through the same API. Which models participate depends on your input and horizon.
Price forecasting with Pretick
AI price forecasts
through a simple API.
Send historical prices and receive a predicted next set of values. Pretick handles the time-series preparation and model calls, so you can add a forecast to an existing chart, market dashboard or application.
What can you forecast?
Use your own chronological observations of stock, cryptocurrency, commodity or other prices. The API forecasts the numerical series you submit. It does not fetch market data, read news or interpret an asset symbol on its own.
A price forecast gives you a baseline to compare with new observations or market expectations. In our gas-price prediction-market case study, an app uses GasBuddy observations to project the remaining price path and adds its own logic to relate that path to AAA settlement prices.
Start with prices. Add timestamps when you have them.
A list of numbers is enough for a first request. Timestamp/value pairs make the timing explicit: Pretick can infer a regular interval and return future timestamps alongside the predictions. You can choose the number of future points or specify a timestamped forecast window.
The example below uses six illustrative prices, sampled every five minutes, and requests three future values. Use enough real history for your application; this short example demonstrates the format.
POST https://pretick.ai/v1/timeseries/forecast
Authorization: Bearer YOUR_PRETICK_KEY
Content-Type: application/json
{
"data": {
"timestamps": [
"2026-09-28T12:00:00Z",
"2026-09-28T12:05:00Z",
"2026-09-28T12:10:00Z",
"2026-09-28T12:15:00Z",
"2026-09-28T12:20:00Z",
"2026-09-28T12:25:00Z"
],
"values": [
3.41,
3.42,
3.4,
3.43,
3.44,
3.42
]
},
"horizon": 3
}Download the exact JSON request. For values alone, the request can be as short as {"data":[3.41,3.42,3.40],"horizon":3}. The default model is consensus.
Read the prediction first, then inspect the detail.
predictions contains the best-guess values. Pair each entry with the corresponding keys entry: an ISO UTC timestamp when dates were supplied, or a sequence number otherwise. The example returns forecast slots at 12:30, 12:35 and 12:40 UTC.
The response also includes the participating model opinions and any model exclusions. You can display one combined line, offer individual model comparisons, or copy the predicted numbers. Consensus uses the median of participating point predictions; the spread between them is model disagreement, not a calibrated probability band.
Choose a model for the data you have.
A financial model opinion
Kronos accepts a single positive price series through its hosted fallback, or supplied open, high, low and close data. It requires continuous history and supports up to 512 observations and 144 future steps.
Handle richer histories without changing providers.
The same endpoint accepts multiple numerical columns, selected prediction targets, and related historical measurements where the model supports them. Separate historical sessions can represent comparable days, with today's observations included as current context.
Timezones, market closures and missing intervals need explicit attention. Pretick supports documented weekday schedules and explicit resampling; it does not silently fill missing observations or infer exchange holidays. Use the free preparation endpoint to inspect the inferred timing before forecasting.
Timestamp handling, historical sessions and input limits
Update the forecast as prices change.
Send another request with the latest observations to obtain a revised path. Your app decides when to refresh and caches the result for display. Each request uses the pretrained model with new context; it does not retrain the model.
Cost and integration
A trial includes five forecasts. Each successful request then costs $0.02, including participating model opinions. Credit is prepaid, with no subscription. Input validation and preparation are free. See forecasting API pricing and the authentication and response guide.
Allow a longer client timeout for financial forecasts: Kronos can take 1–2 minutes. Check usage before retrying a timeout, because a successful request is separately billable. Agents can use the same account through MCP forecasting tools.
Evaluate it against what happens next.
Compare saved forecasts with later prices and a simple baseline such as the last observed value. Match the model, history window and horizon to your actual task. A forecast error score alone does not establish a trading advantage; our financial forecasting research guide explains how to read the available evidence.