What is mantid?
mantid is a Mac app for training machine-learning models and custom language models on data and documents you already have. It runs training on your own Apple Silicon Mac, without an account, and can work with spreadsheets, exports, PDFs, Word files, emails, notes and folders. You can use trained models with new data, ask questions about your documents, or export models for use in your own projects. It is aimed at people and teams who want to explore what their data can answer.
mantid Features:
Local Training: Train predictive and forecasting ML models, as well as custom LLMs, on your own Mac, with your data staying on the computer.
Data Interpretation: Add a spreadsheet, export or folder of PDFs, and mantid interprets the material and indicates what it may be able to answer.
Question Suggestions: Review suggested questions based on your data or enter your own question to guide the model you want to train.
Model Evaluation: Each model is tested against the simplest guess for the same data, and the verdict explains whether the model did better.
Versioned Retraining: Retrain after adding data, correcting a mistake or setting a rule, while keeping earlier model versions available to revisit.
Document Answers: Ask a trained LLM about your files and see which file supports each answer; when the answer is not in the documents, it says so.
Export and Connect: Use models with new data and document folders, export them for your own projects, or call them from AI tools with a local MCP.
mantid Benefits:
Private Workflows: Keeping training on your computer lets you explore company or personal data without sending it to a hosted training service.
Clearer Decisions: Comparing a model with a simple baseline helps you avoid relying on a prediction that does not improve on an easy guess.
Useful Next Steps: A verdict can point to missing data or a needed correction, so you have a concrete way to improve a weak result.
Fewer Document Guesses: Answers tied to their source files make it easier to check what a document-based assistant says and recognize when the material has no answer.
Low-Cost Experimentation: The free app has no account requirement or payment, making it possible to try model training without committing to a paid plan.
Use Cases:
Forecast Bakery Demand: A bakery can use past data to estimate how many croissants to prepare for a Saturday or how many restaurant covers to expect on Friday.
Flag Likely Cancellations: A hotel can examine booking data to identify reservations that may cancel at short notice, while a dental practice can look for likely missed check-ups.
Prioritize Sales Leads: A sales team can train on its existing data to identify which leads may be worth a phone call.
Review Equipment Risk: A workshop can use machine records to predict which equipment may break down, or a factory can investigate which batches may fail inspection.
Search Internal Documents: A team can train on an employee handbook and ask a question such as whether staff can carry unused holiday into the next year.

