An AI tutor for a whole subject. AI gets better, so do you.
Download · Install · First steps · Changelog
Tell Terramentor what you want to learn and by when. The model you connect drafts the course and writes every lesson, diagram and question. Terramentor plans the course against your deadline, asks you to prove each topic rather than just tick it off, and brings back what you are starting to forget.
It needs a model to write with: a hosted one under your own key, or one running on your computer (which to choose). The app itself is free and has no account.
The clip starts from an empty library, with GLM-5.3-Flash as the model. Drafting the course took it two minutes, the first lessons three more, and the tutor's animation about ten seconds. Those waits are cut from the clip.
One loop: Discover → Plan → Study → Prove → Remember. The pictures below come from one course, physics for VWO (the Dutch pre-university exam).
Describe the subject in a sentence and the model drafts a course, as in the clip above: phases,
then topics, each with an overview, in the language you wrote the description in unless you pick
another. The draft runs in the background with an estimate of the time left, so you can keep
studying or start a second course meanwhile. You can also import a course file, a .studyvault
bundle with its media, or an Anki deck. Press c anywhere to drop a link, a paragraph or a photo into
your Inbox.
Give it a start date, a deadline and the days of the week you study. The time is spread across the topics by weight, and pace is measured against that plan, not against the calendar. The dates are the control: change them and whatever is still open is planned again.
The home page is a stream: lessons in short parts, a question after each part, flashcards, and questions from earlier topics you are starting to forget. Diagrams, charts and animations are written by the model as code and checked before you see them. When a course has documents of its own, such as lecture notes or past papers, its lessons are written from them and name the document they used. A tutor for each topic, and an assistant across the whole library, answer from your own files.
A topic is proven when you pass its mastery check (numbers you work out, answers you type, steps you put in order), or when you work an exercise on paper, photograph it and have it marked against a rubric. Before you see a question, the same model answers it again without the key; a key it disputes is thrown out, and a question it could not check is marked on screen and never counts as proof. That catches slips, not a misconception the model holds both times, so a check is only as good as the model behind it. You can always skip a topic, and a skipped topic stays marked as skipped, not as known.
Flashcards are scheduled with FSRS-6, fitted to your own reviews once there are enough of them.
Anki decks come in with their media and review history and go back out as .apkg.
Every topic in your library, placed by what it is about rather than which course it belongs to, on a planet you can turn. Pick a course and it replays the order you finished its topics in, and can save that replay as a video or a GIF.
A chatbot will explain anything you ask, but it keeps no plan, does not know what you have already proved, and never asks you about last month's topic. Anki keeps what you put in it, but you write the cards. NotebookLM answers from your sources and can quiz you on them, but it keeps no deadline and does not bring a topic back when you start to forget it. This app puts those in one loop: the course, the date, the proof, the review.
Lessons and questions are written by the model you connect, pitched at the few lines you wrote about yourself. Switch to a newer model and the next lessons it writes are better, with no update to the app. Lessons already written stay as they are until you regenerate them.
A hosted model is the right choice for most people. The open models an 8 GB graphics card can run write weaker lessons and write them slowly; the ones that teach well want 24 GB or more of video memory. Hosted is also cheap here, because a lesson is written once and kept: the maintainer's own study with a fast, low-cost model (GLM-5.3-Flash on OpenRouter) cost about US$0.20 per day of study, about US$4.40 a month, read from the key's usage on 24 September 2026.
The first screen connects one: Connect OpenRouter signs you in and brings back a key you own, or paste a key for OpenAI or any OpenAI-compatible service, or point it at Ollama, LM Studio or llama.cpp. Without a model the app still opens your courses, their saved lessons and questions, and your flashcards, and writes nothing new.
Your library is one SQLite file on your own disk. There is no account, and nothing reports back to this project. What leaves the machine is what a lesson or an answer needs, sent to the model provider you chose under your own key; and, while the model drafts a new course, search queries written from its topic titles, to find reading for them (on by default, one switch in Settings). Web search inside answers and a daily update check stay off until you turn them on. SECURITY.md lists every outbound connection and how to check the list with a packet capture.
Desktop. Download the zip for your system from the
releases page, unpack it and start
Terramentor.exe (Windows), Terramentor.app (macOS, Apple Silicon) or ./terramentor.sh
(Linux). It carries its own runtime. The launcher is unsigned, so Windows asks once
(More info → Run anyway) and macOS wants a right-click → Open the first time.
Docker, for a server or an always-on machine:
curl -O https://raw.githubusercontent.com/ValeraZSD/terramentor/main/docker-compose.yml
docker compose up -dThen open http://127.0.0.1:3001.
From source (Node.js 22.19 or newer):
git clone https://github.com/ValeraZSD/terramentor.git
cd terramentor
npm install
npm run standaloneUpdates, the Docker password, reaching your library from a phone and what 1.0 promises are in docs/INSTALL.md. The interface comes in twelve languages; English aside, each began as a machine translation, and corrections are welcome (docs/I18N.md).
- First steps: a five-minute tour for a new learner.
- Install and update: desktop, Docker, from source, phone access, upgrades.
- Why it exists: what this is for, and what it is not.
- Architecture: how it is built and why. Start here to contribute.
- Technical reference: schema, engines, API.
- Security: threat model and every outbound connection.
- Contributing, Roadmap, Changelog.
Found a wrong lesson or a bug? Settings → General → About → Report a problem fills in GitHub's form for you to check and send. A factually wrong lesson is one of the most useful reports this project gets.
Copyright © 2026 Valerii Kozhevets (ValeraZSD). AGPL-3.0-or-later, see LICENSE; third-party code is listed in THIRD_PARTY_NOTICES.md. The running app links to its own source under Settings → General → About, as AGPL §13 asks of software used over a network.