wtf is zer0fit?

Zer0Fit is a server that lets an AI chat assistant make predictions from your data without any machine learning training step. It plugs into a chat client through the Model Context Protocol, and once connected you can attach a spreadsheet or CSV file and simply describe what you want predicted in plain language. The system uses two models from Google, TimesFM 2.5 for forecasting future values from historical time series data, and TabFM for classifying rows into categories or predicting numeric outcomes from tables of data. Because these are zero-shot models, there is no need to train anything yourself, tune settings, or engineer features from your columns. In practice, you attach a file in your chat tool, and the assistant automatically figures out the column names, data types, and row counts. Based on the words you use, it decides whether to forecast, classify, or run a regression, then calls the matching tool and returns both the predictions and supporting metrics such as accuracy scores or error rates. For example, you could attach a file of monthly sales and ask for the next twelve months forecast, or attach a dataset of flower measurements and ask it to classify each row by species. Running Zer0Fit requires your own Linux server with an NVIDIA graphics card that has at least 16GB of memory, along with Docker installed. Everything else, including the model code and required libraries, runs inside a Docker container that a single install script builds and starts for you. The project supports both common x86 servers and ARM64 hardware. The README is explicit that this tool is for research and educational use, not for making financial, medical, legal, or safety related decisions, and one of the two underlying models has a non-commercial license. The wrapper code itself is released under the Apache 2.0 license. The project is aimed at people who want to add forecasting or classification abilities to an existing AI assistant setup, particularly those already using Open WebUI, rather than people looking to build machine learning models from scratch.