nixtla
Pre-trained time series foundation model for forecasting and anomaly detection.
Pre-trained time series foundation model for actuarial forecasting and risk anomaly detection.
Nixtla's TimeGPT is a generative pretrained transformer built specifically for time series data. Actuaries and risk analysts can use it to generate zero-shot forecasts and detect anomalies across retail, finance, and operational metrics.
What it is
It provides zero-shot inference, anomaly detection, and fine-tuning capabilities for time series data. You can incorporate exogenous variables like special dates or prices, run multiple series forecasting, and compute prediction intervals.
Who it's for
Suited for quantitative analysts and data scientists needing rapid time series forecasting without manual model tuning. It does not suit teams looking for traditional local-only statistical libraries without external API dependencies.
Key features
Zero-shot Inference
Generate forecasts and detect anomalies out of the box without prior training data.
Exogenous Variables
Incorporate external factors like special dates, events, or pricing changes into predictions.
Snowflake Deployment
Run TimeGPT directly inside your Snowflake environment using stored procedures and UDTFs.
Anomaly Detection
Perform historical and online detection to flag unexpected deviations in risk or claim metrics.
Up and running in 3 steps
Our honest verdict
What's great
- Requires no prior training data for initial zero-shot forecasts
- Supports fine-tuning with custom loss functions and exogenous variables
- Can be deployed directly within Snowflake infrastructure
- Handles irregular timestamps without requiring manual preprocessing
Where it falls short
- Requires obtaining an external API key from Nixtla to execute predictions
- Batch anomaly detection is removed in version 1.0, requiring migration to online methods
- Running inside Snowflake requires specific external access integration setup
Pricing & license
The Python library is free under its open source license, but usage requires an API key obtained through Nixtla's platform, meaning costs depend on your API consumption.
Open remote AI-training roles
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
Does TimeGPT require training data?
No, it features zero-shot inference allowing you to generate forecasts and detect anomalies immediately.
Can I run this inside Snowflake?
Yes, you can install the snowflake extra and deploy forecasting components directly into your database schema.
How do I handle external factors like holidays?
You can add exogenous variables such as special dates, events, or prices to enhance forecast accuracy.
