// the independent publication for AI RPGs
Arcanum is an independent publication covering AI RPGs — the platforms that turn a language model into a real game master. We play every one of them first-hand and score it on a single published rubric, so you can tell which ones hold up over a long campaign, and which quietly fall apart.
Janitor stores a bot as a small set of named fields and counts four of them into every prompt. Most bots fail for mechanical reasons rather than bad writing — they talk for you, they run long, they forget. Here is what each field does, the settings that decide who sees your bot, and the fixes for the three failures people search for most.
Janitor AI's own data model counts four fields into the prompt: personality, scenario, example dialogue and first message. There is no rules field, no state, no dice. So an RPG here is a card that keeps its own books — and every token of ruleset is a token not spent remembering your campaign. Here is how to build one that survives turn thirty.
Android and iOS are not two versions of the same job. On Android you can install the whole SillyTavern server on the phone itself; on iPhone and iPad the project says outright that you cannot, so the phone becomes a browser pointed somewhere else. Here is each route, the commands, and the setting that stops the server booting.
Yes, Character.AI uses water and electricity, and it publishes no figures for either. Here is what can be worked out anyway: the per-message numbers Google and OpenAI have disclosed, the ceiling Character.AI's own cost figure puts on an hour of chat, why estimates differ by a factor of a hundred, and what actually makes roleplay heavier than a quick question.
Chub AI is a front end: it builds the prompt and hands it to a model, either its own or one you connect with a key. So every error lives at one of three places, and Chub's status page only watches two of them. Here is how to tell which one failed, what each error message actually means, and what fixes it.
Master of Dungeon is a solo dark-fantasy RPG for Android and iPhone with real combat, dice, skills and a hero who levels up. It scores 3.3 — the same as Tidefall, reached the opposite way: every axis at 3 or 3.5, the narrowest range we have ever scored. Our review of the rules, the memory, the smithy bug that charged us twice, and the pricing. Rated 3.3/5.
WorldAI (WorldArchitect.AI) is a free D&D 5e game with server-side dice and an engine that holds your items and stats. Our 24-hour story held together and was a pleasure to play. The memory is lossy, the turns are slow, the dice come out too often — and every character in the world shares one opinion of what you did.
// the benchmark
Every AI RPG platform you can actually play, scored on the same seven axes from first-hand play — not from marketing copy, not from a press kit. 25 platforms, one rubric, every score published with its working.
Two of the seven axes have no top score anywhere in the field. Nobody has solved Memory & Continuity and Longevity — not one platform on the board.
| Axis | Field average | Highest reached |
|---|---|---|
| NPC Fidelity | 3.24 | 5 |
| Signature Design | 3.14 | 5 |
| Memory & Continuity | 2.93 | 4.25 · unclaimed |
| Determinism & Fairness | 2.82 | 5 |
| Player Agency | 2.64 | 5 |
| Longevity | 2.62 | 4 · unclaimed |
| Mechanical Depth | 2.52 | 5 |
field average · ceiling reached · outer ring = 5
// how these numbers were made
News, features, guides and interviews — what launched, what changed, and how to get the most out of the games you play.
200 articles →Janitor, Friends & Fables, AI Dungeon and the rest — every roleplay platform played and scored, so you know what's worth your time.
34 platforms → Compare them side by side →The best AI RPGs you can play right now — published GPTs, Gems, and Claude projects — rated and sorted by model and genre.
16 games →Free browser tools for running better campaigns — prompt builder, character cards, context budget, campaign memory. No signup, nothing sent anywhere.
4 tools →// from our workshop
Besides the coverage, Arcanum builds two things: open research on the field, and free tools that run entirely in your browser.
Arcanum RPGs is an independent publication about AI roleplaying games — the platforms and games where a language model runs the story as your game master. We publish news, hands-on reviews, guides and features, and we maintain the Arcanum AI RPG Benchmark, which scores every released platform we have played on the same seven axes. Every rated platform is played first-hand, commercial relationships are disclosed, and scores are never sold.
Arcanum also builds research and tools: the benchmark is published as an open dataset with a permanent DOI, and our free tools help players run better campaigns without sending anything anywhere. Browse the complete AI RPG directory to see every game and platform in one place.