AI Mahjong Tutor is a mahjong coaching agent that teaches tile efficiency, scoring, and table defense across Riichi, American, Chinese Official, Hong Kong, Sichuan, and sanma. Unlike generic chatbots that mix yaku lists into an NMJL Charleston, this agent locks your ruleset first and keeps every discard, score, and drill inside that variant. Built on Jenova's unified platform, it combines leading models from OpenAI, Anthropic, Google, and more with tool integrations for rules research, hand notation, and practice export.
Mahjong Tutor identifies which mahjong you play before it teaches a discard, so Riichi yaku never contaminate NMJL card logic or Hong Kong faan. Yaku, dora, and Mahjong Soul language steer toward Riichi; Charleston, jokers, and exposures toward American; faan toward Hong Kong; the 8-point minimum and 81 patterns toward Chinese Official (MCR). When the signal is unclear, the AI asks rather than guesses — and when 2026 EMA Riichi Competition Rules or a new NMJL card are in force, it can search the current text in-session instead of coaching from last year’s habits.
Mahjong Tutor analyzes the 13 tiles in front of you — MPSZ, Unicode, or plain English — as mentsu and taatsu, not as a shopping list. The path is fixed: confirm variant, round, and seat; map shanten; compare ryanmen waits with kanchan and penchan; count ukeire; then recommend a discard and the principle you can reuse when the next opponent pons. Calls (naki) are weighed as speed bought against value and defense lost, including when a high-scoring direction is too slow for the table.
Mahjong Tutor teaches scoring after the mechanics are solid, building one component at a time against the hand you just built. Riichi players walk han, fu traps such as closed-ron +10 and wait fu, dealer splits, and honba; MCR players work through elements and the five scoring principles; Hong Kong players see faan, minimums, and payment as base × 2^faan. American players learn to read the current NMJL card — categories, notation, joker limits — without receiving a pirated Standard Hands list.
Mahjong Tutor coaches Riichi defense with genbutsu, suji, kabe, and betaori, and American defense with exposure tracking and card elimination. Every variant includes discard reading: suit abandonment, honor timing, and what an open meld announces to three opponents. Ranked players get placement-aware push/fold, including oorasu when a few hundred points separate 3rd from 4th.
Mahjong Tutor places a realistic deal in front of you and waits for your answer before it grades the line. Drills cover discard choice, tenpai identification, defensive suji, and NMJL card reading, then track the pattern across sessions — often sharp waits and soft safe-tile choice. Completed principles can be stored in Google Drive or Notion so the next commute set attacks the same hole instead of restarting at tile names.
| Traditional Approach | Mahjong Tutor |
|---|---|
| Mixed-ruleset videos, unmatched tables | Variant-first Riichi, American, MCR, Hong Kong |
| Nobody reviews your actual discard | Blocks, waits, ukeire, and discard choice |
| Han, fu, faan, and card fog | Step-by-step scoring against real shapes |
| Forgotten leaks by next club night | Persistent variant, level, and pattern memory |
Unlimited Memory vs. Chatbot Amnesia
Mahjong Tutor remembers which ruleset you play, whether you are on the 2026 NMJL card, that you open for speed and then have no safe tile, and that Charleston still freezes you. A generic chatbot that loses that thread will answer a joker-exchange question with Riichi naki logic — an illegal call at an American table and a dead-hand fight. The same amnesia reteaches pinfu while your actual leak is folding after someone declares riichi.
Multi-Model AI Access
You can switch models by phase: a careful stepwise model for a closed tanyao pinfu riichi tsumo han/fu split and non-dealer payment, a stronger table-reading model for dama versus riichi and oorasu placement, and another pass for five legal discard drills with a full pond. ChatGPT is OpenAI-only and Claude is Anthropic-only; this agent draws on OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, and more without juggling accounts. Fu arithmetic and adversarial push/fold are different jobs, and a single-vendor lock-in treats them as the same chat.
Intelligent Tool Integration for End-to-End Workflows
Mahjong Tutor runs a production loop, not a chat about tiles: you paste a hand in MPSZ or English with round, seat, and discards; the AI confirms variant; it looks up a live rule when the 2026 NMJL card or EMA Riichi Competition Rules are in play; it returns block structure, shanten, ukeire, and a discard with a reusable principle; then it drills the leak and stores the note for the next hanchan. The chain is table position → analysis → scored drill → saved pattern — the review that otherwise evaporates between club nights.
Ranked Riichi players
Mahjong Tutor is built for Silver-and-up tables — including Mahjong Soul hanchans — where tanyao, pinfu, and yakuhai are already comfortable and the remaining leak is dealing into obvious ryanmen waits. Paste the deal, name who just declared riichi, and work genbutsu, suji, and placement instead of copying streamer timing.
American mahjong club regulars
Mahjong Tutor helps groups on the 2026 NMJL card who freeze on Charleston, joker exchanges, or exposure legality and want to stop going dead. Card reading, calling conventions, and category announcement are in scope; the copyrighted Standard Hands list is not.
Variant switchers
Mahjong Tutor calls out false friends when you move from Hong Kong faan tables to Riichi, or the other way: a chi that is legal in one ruleset is not how American calling works, and flowers that matter in MCR do not work the same under NMJL. First-week chombo is treated as a curriculum, not a surprise.
Players past four sets and a pair
Mahjong Tutor is for people who know the tiles but cannot yet see waits, ukeire, or when to fold. Lessons follow the order the game unfolds — tiles, melds, turn flow, winning conditions — and delay scoring until you can see a chow.
Commute and tournament players
Mahjong Tutor runs short drill sets and voice-input hand reviews between tonpuusen hands, then carries oorasu notes into the next session. Full parity on phone and desktop means the morning pond and the evening hanchan are the same coaching thread.
Free For exploring Jenova:
Plus (US$20/month) For individuals:
Premium (US$50/month) For creators & freelancers:
Pro (US$100/month) For professionals:
Max (US$200/month) For power users:
Ultra (US$500/month) For unconstrained productivity:
Enterprise (US$1,000/month) For businesses & organizations:
All plans are monthly subscriptions priced in USD, before applicable taxes. Your usage limit resets on your billing date.
Mahjong Tutor is an AI coach that sits with your 13 tiles and trains the next discard, the score, and the defensive fold across Riichi, American NMJL, MCR, Hong Kong, Sichuan Bloody, and sanma. It infers variant and level from how you talk about the game — tile names on one track, shanten and EV on another — then keeps that lens for later sessions. Use it between club nights, ranked hanchans, and tournament weekends when nobody is free to explain why a 5-pin was the wrong fold.
Mahjong Tutor keeps your ruleset, NMJL card year, shanten habits, and defensive leaks in persistent memory, so a Charleston question never receives Riichi naki logic. You can switch among models from OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, and more — a careful model for han/fu and MCR exclusion, a different one for oorasu push/fold, another for legal drill positions. Search, Drive, and Notion close the loop from a pasted hand and a live rule lookup through analysis, drills, and saved notes, which a single-vendor chatbot cannot run as a table-to-review workflow.
Mahjong Tutor uses Google Search to check live targets such as the current NMJL card, 2026 EMA Riichi Competition Rules, and tournament schedules, then folds that into the same session as your hand. Google Drive and Notion hold reviewed deals and leak notes so the next hanchan starts from yesterday’s principle. Speech-to-text on web, iOS, and Android lets you dictate a pond instead of typing MPSZ.
Yes. American (NMJL) mahjong is a first-class variant covering Charleston, jokers, exposures, quints, and card-category strategy. Mahjong Tutor helps you interpret notation, pivots, and calling on the card you bought; it will not paste the copyrighted Standard Hands list. Bring the official 2026 National Mah Jongg League card to the session and ask about Charleston sequence, joker exchanges, and whether an exposure announces your category.
No. Mahjong Tutor locks variant as the default lens before it teaches, and flags Riichi-only ideas as Riichi-only. A chi that is legal in Riichi is not how American calling works; flowers that matter in MCR do not work the same under NMJL. If your first messages are ambiguous, the AI asks which mahjong you play rather than guessing.
Yes. Mahjong Tutor accepts MPSZ notation (123m456p789s11z), 1m–9s style, Unicode tiles, and plain English such as “two green dragons, a 4-5-6 of dots.” Incomplete information gets a clarifying question, not a bluff. For the best read, include variant, round, seat wind, and visible discards or exposures.
No. Mahjong Soul and similar tables are where you play; videos are where you watch someone else play. Mahjong Tutor is where your 13 tiles, your discards, and your variant get a line and a principle. The agent does not replace ranked games or a club teacher; it fills the gap between hands.
Mahjong Tutor treats Riichi han/fu, Hong Kong faan, and MCR’s 81 elements as core expertise and will not invent a yaku, a faan value, or an MCR element. Live targets — a new NMJL card, 2026 EMA Riichi Competition Rules, platform quirks — are checked rather than memorized from last year. If an edge case is uncertain, verify against World Riichi, EMA, WMA, or NMJL rather than treating a guess as a ruling.
Yes. Mahjong Tutor runs with full feature parity on web, iOS, and Android, including speech-to-text, so a suji drill or a dictated hand works on the train after club night. Settings and history sync, so a morning tonpuusen review continues on desktop that evening. Get the same coaching in the Jenova iOS app and the Jenova Android app.
Mahjong Tutor runs discard choice, tenpai identification, defensive suji, wait reading, and NMJL card-reading prompts against realistic visible-tile context. You answer first; then you get the line, the principle, and — across sessions — the pattern, such as sharp tenpai shapes and soft safe-tile choice. Ask for five intermediate Riichi discard problems or a Charleston sequence and the AI waits between items.
Yes. Sichuan Bloody and three-player sanma are supported rulesets alongside Riichi, American, MCR, and Hong Kong. Coaching still starts by locking the variant so Bloody’s missing-suit logic or sanma’s three-player defense is never applied to a four-player Hong Kong table. Level still tracks how you speak — tile names versus shanten and EV.
Mahjong Tutor looks up current governing text when rules moved, including 2026 EMA Riichi Competition Rules and a new annual NMJL card, instead of recycling a static explanation. It will not invent illegal yaku or card hands to fill a gap. When the edge case remains uncertain, it tells you to verify with the governing body rather than bluff a ruling.
Mahjong Tutor is part of Jenova, the AI agent platform for creativity, entertainment, and life. Jenova provides unified access to leading AI models from OpenAI, Anthropic, Google, xAI, DeepSeek, and more — alongside intelligent tool integrations for end-to-end task completion. The platform features persistent cross-session memory and runs on web, iOS, and Android, with your agents and settings synced across devices. Developed by Azeroth Inc., a New York-based technology company, Jenova is used by people around the world.
For developers, the Jenova Agent API provides a fully-managed agent backend with built-in orchestration, multi-model routing, memory, storage, and MCP tool support — replacing the need to stitch together raw LLM APIs, vector databases, and custom tool servers.
Contact: [email protected]