Repolode
Submit
Nixtla

Nixtla/nixtla

Python SDK client for Nixtla's TimeGPT, a hosted pre-trained foundation model for time series forecasting and anomaly detection.

View on GitHubWebsite
Stars
4,026
Forks
341
Open issues
63
Last commit
4 days ago

What is it?

This repository contains the Nixtla Python SDK, a client for TimeGPT, a generative pretrained transformer for time series. It lets developers forecast future values and detect anomalies in time series data with a few lines of code, without training a model on their own data first.

TimeGPT itself is closed source and runs on Nixtla's servers, so every forecast or anomaly detection call requires an API key. The SDK is open source under Apache 2.0. It can also be used with TimeGPT deployed on your own infrastructure, and includes a script for deploying TimeGPT inside Snowflake.

The SDK supports zero-shot forecasting, fine-tuning, exogenous variables, prediction intervals, cross-validation, and multi-series forecasting. Its anomaly detection API is batch-based in the current version, and the README notes that a different online variant is planned for nixtla 1.0.

Key features

  • 01Zero-shot forecasting and anomaly detection without training a model
  • 02Fine-tuning on your own datasets with custom loss functions
  • 03Support for exogenous variables such as holidays, events, or prices
  • 04Prediction intervals and built-in cross-validation
  • 05Multi-series forecasting and irregular timestamps
  • 06Snowflake deployment script for running inside your data warehouse

Use cases

  • Forecasting retail demand or electricity load without building custom models
  • Monitoring IoT sensor or metric streams for anomalies
  • Adding time series forecasts to an application via API calls
  • Running forecasts inside Snowflake without moving data out

Quick start

  1. 1

    Install the nixtla SDK

    pip install nixtla>=0.8.0
  2. 2

    Get an API key from the free trial page

  3. 3

    Create a NixtlaClient with your key and call forecast() on your data

Things to know

  • An API key is required, since forecasts run on the hosted TimeGPT model.
  • TimeGPT itself is closed source; only the SDK client is Apache 2.0 licensed.
  • detect_anomalies() is removed in nixtla 1.0 and has no drop-in replacement, so upgrading will break batch anomaly detection code.

Topics

#time-series#time-series-forecasting#anomaly-detection#foundation-models#python-sdk#forecasting#agent#agentic-ai#artificial-intelligence#deep-learning

FAQ

What is nixtla?
Python SDK client for Nixtla's TimeGPT, a hosted pre-trained foundation model for time series forecasting and anomaly detection. This repository contains the Nixtla Python SDK, a client for TimeGPT, a generative pretrained transformer for time series. It lets developers forecast future values and detect anomalies in time series data with a few lines of code, without training a model on their own data first.
Is nixtla free to use? What license does it have?
nixtla is open source under the Other license. Check the license terms before using it in commercial products.
Is nixtla still maintained?
Actively maintained. The last commit to nixtla was on Oct 15, 2026.
What can I use nixtla for?
Forecasting retail demand or electricity load without building custom models Monitoring IoT sensor or metric streams for anomalies Adding time series forecasts to an application via API calls Running forecasts inside Snowflake without moving data out
How do I install nixtla?
Install the nixtla SDK: pip install nixtla>=0.8.0 Get an API key from the free trial page Create a NixtlaClient with your key and call forecast() on your data
Where can I get nixtla?
The source code and installation instructions are on GitHub: https://github.com/Nixtla/nixtla

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects