The AI RPG Directory (2026): Every LLM RPG Game & Platform for ChatGPT, Claude, Gemini & Grok
If you search “AI RPG” in 2026, you find the same thing on every page: a platform selling you its own product while pretending to give you a guide. Jenova wants you to subscribe to Roleplay Game Master. HowWorks wants your click. Wanderfolk put themselves first on their own list. AI Dungeon reviewed AI Dungeon.
Arcanum has no product to sell you and no subscription to push. This is an AI RPG directory — a curated, independent hub for discovering AI roleplaying games and platforms, whether you searched for that or for “AI storytelling games” — with a focus on LLM-native roleplay games that run inside the AI you already use. No install. No new subscription. No separate platform. Just games engineered to run inside ChatGPT, Claude, and Gemini, reviewed and rated first-hand by someone who has spent four decades studying how roleplaying systems work.
That is the only lens this directory uses. Does it hold together? Does it respect your agency? Does it solve the problems that kill most AI campaigns? Does it give you a game worth playing — or just a prompt that breaks down after twenty turns?
What Is an LLM RPG Game?
Before the directory, a necessary distinction that most content in this space blurs deliberately.
LLM-native RPG games are designed to run directly inside a large language model interface — a Custom GPT inside ChatGPT, a Project inside Claude, a Gem inside Gemini. They require no separate platform, no additional subscription, and no install. The game engine is a master prompt, sometimes paired with a knowledge file. You bring the AI you already use; the game runs inside it.
AI RPG platforms are standalone products — Friends & Fables, AI Dungeon, Jenova, Voyage, Janitor AI. They require their own accounts, their own subscriptions, and their own interfaces. The AI is built in; you bring nothing but yourself.
Both are valid. They serve different needs. This directory starts with the first category: games you can play today inside ChatGPT, Claude, or Gemini without paying anything extra.
For the platform category, Arcanum maintains separate, honest reviews at /clients/.
How Arcanum Evaluates LLM RPG Games
Every game in this directory is assessed against the same criteria. These are not arbitrary — they reflect the actual failure modes of LLM campaigns, developed over years of playing and testing games in this specific medium.
Memory architecture. How does the game handle context window limits? Does it have compression systems, save-state protocols, or structured player logs built into its design? A game with no answer to the turn-50 amnesia problem is not a long-campaign game, whatever it claims.
Player agency protocol. Does the game genuinely protect the player’s right to decide? The most common failure in LLM games is narration drift — the AI begins making decisions for the player, inventing their dialogue, fast-forwarding through scenes the player should be playing. Well-engineered games build explicit, hard-walled agency protection into their prompt architecture.
World and mechanical depth. Does the game have genuine systems — economy, faction reputation, relationship dynamics, tactical resolution — or is it a thin aesthetic wrapper over a generic roleplay prompt? Depth is what makes a game replayable rather than exhaustible.
Deterministic or probabilistic resolution. Does the game lean on dice rolls and random chance, or does it resolve outcomes through the quality of the player’s preparation and decisions? The best LLM games reward intelligence, not luck.
Signature design. What does this game do that nothing else does? A game without a distinctive mechanical identity is indistinguishable from a bare prompt.
Longevity. Is this a one-session experience or a campaign engine? Can the player run it for fifty hours and still be discovering new things?
Ratings are out of 5. No game in this directory receives a rating for features it claims but does not deliver.
Public LLM RPG Games We Have Reviewed
These are community and studio games that run inside ChatGPT, Claude, or Gemini, each played and scored by Arcanum against the same published methodology as the platforms. Scores are out of 5.
| Game and score | Runs on | Best for |
|---|---|---|
| Solo RPG Master 2.6/5 | ChatGPT | Multi-universe solo play with standout characters and romance |
| Deep Saga 2.4/5 | ChatGPT | Dangerous, illustrated sagas across fantasy, sci-fi and horror |
| Isekai RPG 2.4/5 | ChatGPT | An original isekai world with deep NPCs and beast taming |
| Vantiel 2.4/5 | ChatGPT | Rich lore and polished presentation |
| 8-Bit Kingdoms 2.3/5 | ChatGPT | Generational dynasty management |
| Valkyrie’s Biggest Gig 2.3/5 | Claude | Cyberpunk where your body is the resource you spend |
| Burning Sun V2 2.1/5 | Gemini | A world that reacts to what you do |
| Gamekeeper RPG Prompt 2.1/5 | Gemini | A visible JSON save-state |
| RPG GPT 2/5 | ChatGPT | A GM that stays in character |
| Classic Text Adventure Prompt 1.9/5 | Claude | An old-school room-by-room adventure |
| Realm & Companion RPG Prompt 1.9/5 | Gemini | Four-choice turns and alignment tracking |
| Solo D&D Starter Prompt 1.6/5 | Claude | The fastest possible start |
This section is regularly updated as new games are reviewed. Submit a game for consideration via arcanumrpgs.com.
For full write-ups and filters by model and genre, visit the Games Directory →
Made by Arcanum, not ranked: the four Arcanum Originals are prompt games we made ourselves. We never score or rank our own work, so they sit outside this list.
The AI RPG Platform Landscape
LLM-native games are one half of the AI RPG ecosystem. For players who want a dedicated platform with its own interface, visual tools, or a built-in community, the following have been independently reviewed by Arcanum.
| Platform | Rating | Pricing | Best For |
|---|---|---|---|
| Friends & Fables | 3.9/5 | Freemium | Visual tabletop, D&D 5e tactical combat, group play |
| RoleForge | Closed alpha (unrated) | Free | Solo campaigns, clean narrative interface, active development |
| Deep Realms | 2.7/5 | Freemium | Fantasy world-building, rich NPC systems |
| Janitor AI | 3.0/5 | Freemium | Largest character library, BYOM flexibility |
| MacerAI | 2.9/5 | Freemium | Character-driven roleplay, strong persona consistency |
| Taverna | Closed beta (unrated) | Freemium | Collaborative play, community-built worlds |
| Voyage | 4.4/5 | Freemium | Structured successor to AI Dungeon: real dice, hard combat, permadeath |
| AI Realm | 2.3/5 | Freemium | Accessible entry point, broad genre support |
| Infinity DM | 2.7/5 | Freemium | DM assistant tools, tabletop prep |
| AI Dungeon | 2.6/5 | Freemium | Maximum creative freedom, the original AI RPG |
All platform reviews are independent and scores are never for sale. Where we received free access to a platform, its review says so. Full reviews at /clients/.
The wider landscape. The category is bigger than the platforms Arcanum has tested first-hand. Others worth knowing — listed here as pointers rather than rated entries, because a rating requires a genuine first-hand review — include Saga (long-form persistent-world AI text RPG), aiga_ (an AI story game with naturally branching narratives), Master of Lore (an AI game master for fantasy and myth campaigns), and LitRPG Adventures (AI worldbuilding and generator tools for GMs). As each is reviewed, it moves up into the rated directory above.
Choosing the Right Model for LLM RPG Play
The model you play on matters. Each of the major consumer LLMs has genuine strengths and real limitations for roleplay specifically.
ChatGPT (Custom GPTs) The most feature-rich environment for LLM gaming. Custom GPTs allow a dedicated game container — separate from your regular ChatGPT history, with a persistent system prompt and optional knowledge files. Excellent at complex world-building, logical consistency, and rule interpretation. The Custom GPT store also has the largest library of community-built RPG games of any platform. Best suited to games with explicit mechanical systems where logical consistency under rules pressure matters. For curated picks in this category, see the best ChatGPT RPG games →.
Claude (Projects) The strongest prose writer of the three and the most reliable at maintaining character voice and emotional nuance across a long conversation. Claude Projects work similarly to Custom GPTs — a persistent system prompt, uploaded knowledge files, and a contained game environment. The most natural fit for relationship-heavy games where the quality of NPC dialogue and psychological depth matter more than mechanical precision. Claude also enforces player agency more naturally than the other models, making it a stronger host for games with strict agency protocols. For the standouts, see the best Claude RPGs →.
Gemini (Gems) The practical advantage Gemini has over the other two is context window: Gemini’s 1M-token window on Google AI Pro is one of the largest available to consumer users, which means longer campaigns before memory compression becomes critical. Gemini Gems function like Custom GPTs and Claude Projects — a dedicated game container with a persistent prompt and knowledge files. Well-suited to games with large world bibles where the model needs to hold a lot of lore in working memory simultaneously. The Gem library is currently the thinnest of the three, which is both a limitation and an opportunity for early content. See the current standouts in the best Gemini RPG games →.
Grok (Custom Agents & Workspaces) The newest of the group and the strongest on emotional intelligence and permissiveness. Grok leads roleplay emotional-IQ rankings and refuses dark or mature scenes far less than the others, making it a natural fit for character-driven, companion, and freeform play. Its container is a pair — Custom Agents (named GM personas) plus Workspaces (persistent files and instructions) — and the fast tier’s multi-million-token context suits very long sessions. The trade-off is weaker rule-following than Claude or GPT, so it’s a better character model than a strict-mechanics engine. See the Grok RPG guide → or grab two ready-to-paste Grok RPG prompts →.
For a full comparison including DeepSeek, MiniMax M2, and local models, see Best LLM for Roleplay →.
The Problems That Kill Most AI Campaigns (And How Good Games Solve Them)
Most LLM RPGs fail not because of bad writing but because they don’t solve the structural problems of the medium. Understanding these failures helps you evaluate any game you encounter.
The turn-50 amnesia problem. Every LLM has a context window — a limit to how much conversation it holds in active memory. When a campaign exceeds that window, the oldest information begins dropping. The NPC who swore a blood oath in session three greets you like a stranger in session ten. Character names drift. Established world rules get forgotten. Well-engineered games build explicit memory compression systems into their prompt architecture — structured save-states, campaign memory protocols, or player-maintained logs that reinstate context at the start of each session. Games without this are not long-campaign games. See our Campaign Memory Tool →
Narration drift. The most common agency failure in LLM games: the AI begins making decisions for the player. It narrates their thoughts. It invents their dialogue. It fast-forwards through scenes the player should be playing. This typically starts small and compounds over a session. Well-engineered games include explicit, hard-walled agency protection in their prompt — the equivalent of a constitutional rule that the player controls all thoughts, words, and actions, with a specific command to invoke when the model drifts. Games without this will consistently override the player’s choices.
The thin-prompt collapse. Many “AI RPG games” are not games at all — they are a single paragraph of setup with no mechanical depth, no memory architecture, and no systematic world. They work for one session and collapse into generic fantasy storytelling by the third. A genuine LLM RPG game has a world canon, a resolution system, a relationship engine, and explicit rules for the AI to follow. The craft is in the prompt engineering, and the difference between a good game and a thin one is immediately apparent to anyone who has played both.
Random chance substituting for preparation. Many LLM games lean on dice rolls as the primary resolution mechanic. This is a design shortcut — it means the game doesn’t need to model the quality of the player’s preparation or the intelligence of their approach. The best LLM games are deterministic: outcomes come from what the player knows, what they’ve prepared, and how well they’ve thought through their approach. Luck is not a design philosophy; it’s an absence of one.
Frequently Asked Questions
What is an AI RPG directory? An AI RPG directory is a curated hub for discovering AI-driven roleplaying games, game-master tools, and platforms. A good one catalogs both LLM-native games — Custom GPTs, Claude Projects, and Gemini Gems that run inside an AI you already use — and standalone AI RPG platforms like AI Dungeon or Friends & Fables, with independent reviews rather than self-promotion. Arcanum is a dedicated AI RPG directory covering both, with first-hand ratings and no paid listings.
What is the difference between an LLM RPG game and an AI RPG platform? An LLM RPG game runs inside an AI you already use — a Custom GPT inside ChatGPT, a Project inside Claude, a Gem inside Gemini. No separate account, no new subscription. An AI RPG platform is a standalone product with its own interface, account, and usually its own pricing. Both have value; they serve different preferences. This directory covers the first. Arcanum’s clients section covers the second.
Do I need a paid subscription to play these games? Most of the games in this directory run on the free tiers of ChatGPT, Claude, and Gemini, though free tiers have usage limits. Games with large knowledge files may benefit from paid tiers that allow larger file uploads.
How is this different from just prompting ChatGPT to run a fantasy RPG? A bare prompt produces a generic session that collapses within twenty turns. Engineered LLM RPG games have explicit memory architectures, player agency protocols, deterministic resolution systems, world canons, faction simulators, relationship engines, and signature mechanical identities. The craft gap between a thoughtful bare prompt and a well-engineered game engine is enormous — approximately equivalent to the gap between a rough sketch and a finished design.
How often is this directory updated? The public games and platform sections are updated as new reviews are completed. If you’ve built or found an LLM RPG game that should be in this directory, contact Arcanum at arcanumrpgs.com.
What’s the best LLM RPG for a complete beginner? Solo RPG Master is the easiest first session: a multi-genre solo storyteller you open in the ChatGPT store, with no files to load. On Claude, Valkyrie’s Biggest Gig is a single copy-paste prompt, and on Gemini, Burning Sun V2 is the same.
What’s the best free AI RPG game overall? Of the public games we have reviewed, Solo RPG Master scores highest, and it opens straight from the ChatGPT store. For free platforms and free tiers, see the best free AI roleplay guide.
Can I play these on mobile? Yes. ChatGPT, Claude, and Gemini all have mobile apps. Custom GPTs, Projects, and Gems are accessible on mobile. The games in this directory are fully playable on phone or tablet.
What does “LLM-native” mean? A game designed from the ground up for a large language model interface, rather than adapted from a traditional tabletop or video game framework. LLM-native design accounts for the specific strengths and failure modes of language models — context windows, character drift, agency erosion, resolution without dice — in a way that traditional game design frameworks do not.
About This Directory
Arcanum RPGs is an independent publication covering AI roleplaying games. Every review is written from first-hand play by Rukka Nova, whose background in roleplaying spans four decades across tabletop systems, computer RPGs, MUDs, and emergent simulation games.
No game or platform can pay for a listing, a review, or a rating. Arcanum does sell clearly labeled sponsored placements, only to platforms that are already listed, and design services to companies in this market; neither touches a score, and any commercial relationship or free access is disclosed on the pieces it affects. The rules are in our editorial policy.
The directory grows as the category grows. If you have built an LLM RPG game that belongs here, or found one that does, Arcanum wants to know about it.