{"slug": "show-hn-mantid-a-mac-app-that-lets-non-ml-people-train-their-own-ai-models", "title": "Show HN: Mantid, a Mac app that lets non-ML people train their own AI models", "summary": "A Mac app called Mantid lets non-ML users train custom ML models and LLMs locally on their own data, with no account and no cost, according to the app's Show HN launch post. Mantid requires a Mac with Apple Silicon (M1 or later) and macOS 14 or later, with 8 GB of memory for ML models and 16 GB for LLMs, and it benchmarks every trained model against the simplest possible guess, reporting whether it beat that baseline. The app exports models for DIY deployment and exposes them to AI tools through a local MCP, and the post states user data never leaves the machine, though anonymous usage data is on by default and can be turned off.", "body_md": "# Vibe train AI models\n\nFeed mantid the data and files you already have. Train an ML model that predicts and forecasts, or train a custom LLM for general AI.\n\nFree. No account required. Your data never leaves your computer.\n\n## Drop in your data. mantid trains the ML model.\n\n### Bring what you have.\n\nA spreadsheet, an export, a folder of PDFs. mantid reads it and says what it can do with it.\n\n### Pick a question.\n\nmantid suggests questions your data can answer (or type your own).\n\n### Go and make a cup of tea.\n\nIt trains on your machine while you relax. mantids are patient.\n\n### Get a verdict.\n\nmantid reports back on the custom AI model.\n\n### Put it to work.\n\nDrop in next week's data to get answers, or export the model into your own project.\n\n| weekday | days_ahead | past_no_shows | sms_sent | no_show | \n|---|---|---|---|---|\n| Mon | 21 | 2 | no | yes | \n| Tue | 3 | 0 | yes | no | \n| Tue | 14 | 1 | yes | no | \n| Wed | 30 | 3 | no | yes | \n| Thu | 1 | 0 | yes | no | \n\n### What do you want to know?\n\n## Bad models get eaten.\n\nEvery model is measured against the simplest guess anyone could make from the same data. mantid tells you whether it beat that guess, and what would help if it didn't.\n\n### Retrain your models with versioning.\n\nNothing is overwritten. Go back to any version you liked better.\n\n### Every verdict comes with a fix.\n\nAdd data, correct a mistake, set a rule, then run it again.\n\n### This model doesn't beat the simple guess.\n\nThe simple guess says nobody leaves. It is right 97 times in 100. Your model is right 96.\n\nWhat would help: only 38 of your 1,200 customers left, so add a column like last_login and bring two years of customers instead of one.\n\n## Bring fuzzy data. Let mantid figure it out.\n\nmantid interprets your data for you.\n\nGyms\n\n### Which gym members are about to ghost us?\n\ncancelled\nDental practices\n\n### Who's going to skip their check-up?\n\nno_show\nBakeries\n\n### How many croissants do we bake on Saturday?\n\nunits_sold\nHotels\n\n### Which bookings will cancel last minute?\n\ncancelled\nWorkshops\n\n### Which machine breaks down next?\n\nfailed_30d\nFactories\n\n### Which batches fail inspection?\n\nfailed_qa\nSoftware teams\n\n### Which trial users will actually pay?\n\nconverted\nSales teams\n\n### Which leads are worth a phone call?\n\nwon\nRestaurants\n\n### How many covers on Friday night?\n\ncovers\nCharities\n\n### Which donors will give again this year?\n\ngave_again\nSupport teams\n\n### Which support emails are actually urgent?\n\nurgent\n### Your oddly specific question goes here\n\nIf you've got the data, mantid will tell you if it can be answered.\n\n## Train it on your documents. Then ask it anything.\n\nDrop in your documents and data. mantid trains your own LLM on them, then you use it like any assistant, on your own machine.\n\n### Give it anything\n\nPDFs, Word files, emails, notes and spreadsheets. As much as you have.\n\n### Ask your documents\n\nEvery answer from your files shows the file it came from, and \"not from your documents\" beats a guess.\n\n### Put it to work\n\nTell it what to do, hand it a new file to read, and get text or a table back.\n\n### Tested before you use it\n\nmantid checks it against the untrained model on questions from your own files it never saw.\n\nCan staff carry over unused holiday?\n\nMany companies let staff carry over some holiday. Check your employee handbook or ask HR.\n\nYes, up to five days into the next year, if your manager agrees in writing before 31 December.\n\nHoliday policy.pdf, p.3\n## Pricing. Nom nom.\n\n### Free\n\nYou read that right.\n\nTrain as much as you like on your own machine. Nothing to pay and no account.\n\n- Train as many ML models and LLMs as you like on your data.\n- Use them on new data, files and folders of documents.\n- Export your models for DIY deployment.\n- Call your models in your AI tools with a local MCP.\n- Your data never leaves your machine.\n\nA Mac with Apple Silicon (M1 or later), macOS 14 or later. ML models need 8 GB of memory, LLMs need 16 GB. Anonymous usage data is on by default and can be switched off.\n\n### Hosted with API & MCP\n\nUse your models in any platform, AI tool or vibe coded app. High uptime, low cost, usage based fees.", "url": "https://wpnews.pro/news/show-hn-mantid-a-mac-app-that-lets-non-ml-people-train-their-own-ai-models", "canonical_source": "https://mantid.dev", "published_at": "2026-10-24 12:49:06+00:00", "updated_at": "2026-10-24 13:11:13.634446+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-tools", "agent-protocols"], "entities": ["Mantid", "Apple Silicon", "macOS 14", "MCP"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/show-hn-mantid-a-mac-app-that-lets-non-ml-people-train-their-own-ai-models", "markdown": "https://wpnews.pro/news/show-hn-mantid-a-mac-app-that-lets-non-ml-people-train-their-own-ai-models.md", "text": "https://wpnews.pro/news/show-hn-mantid-a-mac-app-that-lets-non-ml-people-train-their-own-ai-models.txt", "jsonld": "https://wpnews.pro/news/show-hn-mantid-a-mac-app-that-lets-non-ml-people-train-their-own-ai-models.jsonld"}}