Show HN: TimeCopilot, forecasting agent with LLMs and foundation models (github.com)

TimeCopilot is an open-source forecasting agent that innovatively integrates large language models (LLMs) with advanced time series foundation models like Amazon Chronos, Salesforce Moirai, and Google TimesFM. It automates complex forecasting workflows by interpreting statistical features, guiding model selection, and explaining results in natural language. This approach enhances accessibility to time series analysis without compromising professional-grade accuracy, making sophisticated forecasting techniques more approachable for practitioners and researchers alike. Technically, TimeCopilot supports seamless, one-command forecasting directly from public datasets via a command-line interface or Python API, leveraging LLMs such as GPT-4 for interpretability and interactivity. It provides detailed outputs including feature analysis, model selection rationale, cross-validation results, and natural language responses to domain-specific queries about forecasts. For instance, it can determine the best model automatically (e.g., AutoARIMA for trend and seasonality), generate predictions, and answer questions like projected total passengers over a period, trend growth rates, or seasonal peaks. This hybrid LLM-foundation model strategy facilitates transparent, explainable forecasting and enables users to query forecasts in natural language, a meaningful leap for AI-driven time series analysis workflows. By combining robust statistical modeling with generative AI’s interpretative power, TimeCopilot promises to streamline forecasting practices, reduce reliance on specialized expertise, and foster deeper insights across industries. Its open-source nature and community-driven development also invite collaboration and continuous enhancement, positioning it as a significant tool for the AI/ML community invested in time series prediction and explainability.