Timecopilot

TimeCopilot is an open source forecasting agent that pairs large language models with time series foundation models. It selects and runs forecasting models, explains its choices in natural language and answers questions about the results. The vendor lists more than 30 time series foundation models in its unified layer, including Amazon Chronos, Salesforce Moirai, Google TimesFM and Nixtla TimeGPT. Users interact in plain English rather than through custom pipelines. The agent interprets statistical features of a dataset, guides model selection from the data characteristics and explains the technical decisions it made. A single command line call, run through uvx, can forecast a public dataset from a URL, and the user can choose the language model and pass a specific question about the forecast. The documented default model is OpenAI's GPT 4o mini. The package is distributed on PyPI for Python development, and the documentation covers examples, fine tuning of foundation models, benchmarks and ensembles, and integrations. It is aimed at analysts, data scientists and engineers who want forecasting that is more accessible without giving up accuracy. The project reports a top position on the GIFT Eval benchmark and an accepted paper at a NeurIPS workshop. The site publishes no pricing; the software is open source, and language model usage depends on the provider the user configures.