TokenMax — Learn AI & Python

TokenMax is an AI learning lab for your iPhone, built by Vellora AI: a full computer-science curriculum, a real Python environment with auto-graded problem sets, a sandboxed terminal, and a tutor powered by a language model that runs on the phone itself. No accounts, no servers, no telemetry. LEARN AI, CS & PYTHON — 14 FULL COURSES, OFFLINE - Sidebar → Courses bundles 14 university-level open-courseware classes, readable offline, including: • 6.S191 Intro to Deep Learning — transformers, attention, RLHF, generative AI • 6.5940 TinyML & Efficient Deep Learning — quantization, pruning, the methods behind on-device models • 6.0001 + 6.0002 Intro to CS and Data Science in Python • 6.034 Artificial Intelligence, 6.864 Advanced NLP • 18.06 Linear Algebra, 18.065 Matrix Methods, 18.05 Probability & Statistics • 6.006 + 6.046 Algorithms, 6.S081 Operating Systems, 6.824 Distributed Systems - Course materials from MIT OpenCourseWare, used under CC BY-NC-SA 4.0. TokenMax is free and is not affiliated with or endorsed by MIT. Lecture videos link out to ocw.mit.edu. A REAL PYTHON LAB IN YOUR POCKET - Embedded Python 3.13 — write code and run it without leaving the app. - Auto-graded problem sets check your work line by line, with results in seconds. - Built for learning: the interpreter runs only the bundled lessons and the code you type. It never downloads or executes code from the internet, and network access inside the Python sandbox is disabled. - "Python Tutor (OCW 6.0001)" persona — a Socratic tutor that asks leading questions, runs your code through Python before commenting, and refuses to just hand you the answer. TERMINAL & SSH - A real sandboxed shell with twenty-something built-ins (ls, cd, cat, head, tail, grep, find, mkdir, mv, cp, rm, touch, …) operating only on the app's own writable directories. - SSH client in pure Swift. Save hosts in Settings, connect with password or ED25519 / RSA private keys, host-key fingerprint pinning on first connect. - Per-conversation Plan / Execute mode lets the model drive the terminal. Plan mode allows read-only commands; Execute mode requires per-command approval for anything destructive. AN AI THAT NEVER PHONES HOME - A language model ships inside the app — start chatting on first launch, even in Airplane Mode. - Streaming responses, regenerate, edit and re-roll, copy, pin, folders, in-conversation search, export to PDF / Markdown / JSON, share extension. - Sees photos, hears your voice, reads PDFs. Vision-language models run natively on Apple Silicon via mlx-swift: Qwen2.5-VL 7B on iPhone 15 Pro / 16 Pro / 17 Pro, Qwen2-VL 2B on iPhone 13+. Attach a photo while a text-only model is active and TokenMax reads it on-device with OCR instead. - Drop a PDF, .docx, .md or .txt and ask questions about it — retrieval runs locally with Apple's NLEmbedding. - Web search with real source citations, only when you ask for it. - Slash commands: /summarize, /explain, /code, /tldr, /translate, /improve. PRIVACY YOU CAN AUDIT - Your work stays on your iPhone. The only network calls TokenMax can make are optional model downloads, web search you trigger, and SSH connections to servers you've added — and Settings → Privacy & Data lists every one of them: the host, what data leaves, and what triggers it. - Sealed mode hard-blocks all network traffic with one switch. - Versioned in-app Terms and Privacy Policy. REQUIREMENTS - iPhone 13 or newer recommended. Runs on iPad too. iPhone 15 Pro / 16 Pro / 17 Pro for the 7B vision model. - Falcon 3 1B ships pre-installed — no download needed to start. Larger open-weight models are optional downloads that always show their exact size first (Qwen 2-VL / 2.5-VL under Tongyi Qianwen License, Falcon 3 under TII Falcon-LLM License 2.0, Bonsai 8B under Apache 2.0). TokenMax is built by Vellora AI. We don't run a server. We can't read your work because it never reaches us.