Spavik
Get API key

From historical data to a prediction

Provide data with known outcomes. Spavik creates a model to predict those outcomes for new data.

Create, use and update a model

  1. 01

    Evaluate

    Check whether your history contains enough information to predict the chosen value.

    POST /v1/datasets/evaluate
  2. 02

    Create a model

    Choose the column to predict. Spavik trains a model using your historical data.

    POST /v1/models
  3. 03

    Predict

    Send new rows. The model returns a predicted value for each one.

    POST /v1/models/{id}/predict
  4. 04

    Send outcomes

    Report the values you actually observed.

    POST /v1/models/{id}/outcomes
  5. 05

    Monitor

    Compare predictions with real outcomes.

    GET /v1/models/{id}/performance
  6. 06

    Refresh

    Create a new version that includes the outcomes you reported.

    POST /v1/models/{id}/refresh

Forecast the next values in a time series

Measure forecast accuracy on a known period, then request future values. No persistent model is needed.

  1. BacktestPOST /v1/forecast/backtest
  2. ForecastPOST /v1/forecast

What Spavik handles for you

You choose the column to predict. Spavik does the rest, with no ML knowledge needed. To calculate predictions, it keeps examples from your history as part of the model it hosts.

Model choice
Automatic, based on the column you predict
Settings to tune
None
Categories and missing values
Accepted as they are
Hosting
By Spavik, nothing to deploy
Training
Free
Minimum
10 rows, with at least 2 different values to predict
Updates
A new model version

What data can you use?

Supported data
  • Historical customer, order, transaction or sensor data
  • A prediction per row with predict(), time series forecasts with forecast()
  • Columns containing numbers, categories or missing values
  • At least 10 rows to create a model; this does not guarantee that the data is predictable
Limitations
  • Images, audio and video
  • Free text as the main signal
  • Data only known after the outcome: it distorts the model evaluation
  • A series with fewer observations than the number of periods to forecast

Errors that say what to do next

Every error returns a stable code for your program and a detail that explains what was refused and how to fix it. A developer or an AI agent can act on it without opening the docs.

Error response 
{
  "status": "error",
  "message": "QUOTA_MODELS_EXCEEDED",
  "messageDetail": "Model limit reached (3/3) on the build plan. Archive a model you no longer serve, or upgrade to the pro plan."
}

Start by evaluating your data

Get an API key. Evaluating a table is free.

Get API key Read the docs