Layer 1
AI as the DM
The AI runs the game. It narrates scenes, voices NPCs, adjudicates rules, tracks combat. You are the player, not the GM.
Best for solo play, async groups, players whose table fell apart, and collaborative fiction.
The honest guide
Most "Best AI Tools for D&D" lists are written by an AI tool company that puts itself at #1 and treats "AI that runs the game" and "AI that helps you prep" as the same thing. They aren't. This is the guide organized by what you actually want to do, with honest takes on each tool.
Zach "Zafety" Townes solo dev of Grimoire ~10 min read
The bias is in the byline: I built one of the tools on this list (Grimoire). I tried to write the version that's useful even if you pick a different one.
The frame that makes the rest clear
Once you can see the layers, picking a tool stops being a feature-checkbox exercise and becomes one question: which layer fits the kind of game you want to run?
Layer 1
The AI runs the game. It narrates scenes, voices NPCs, adjudicates rules, tracks combat. You are the player, not the GM.
Best for solo play, async groups, players whose table fell apart, and collaborative fiction.
Layer 2
The AI helps the human GM between sessions: parsing notes, tracking state, generating ideas grounded in your canon, recalling lore, drafting recaps. You run the table; the AI handles the bookkeeping.
Best for human-DMed campaigns where you want less prep overhead and continuity that survives years.
Layer 3
Single-purpose generators that produce specific assets you bring to your own game: NPCs, encounters, magic items, battlemaps, portraits. You pick what to use; the AI makes it.
Best for any DM who wants a faster prep loop but doesn't want the AI involved at the table itself.
You can use tools from more than one layer, and most experienced GMs do. The mistake is buying a Layer 1 tool and expecting it to solve a Layer 2 problem, or the reverse. I go deeper on the Layer 2 philosophy on the AI for Dungeon Masters page.
Why the distinction matters
A character prediction, mid-campaign
How would Queen Mira react to her brother's betrayal?
Layer 1 — AI as the DM
Tries to roleplay Mira on the spot. Reads her as a generic betrayed-queen trope. It has no idea your players already uncovered that she suspects her brother Cyrus is the cult's silent backer, because that fact lives in your notes, not its context. So it gives a plausible, well-written, wrong answer. And with no separate source of truth, that wrong answer can quietly become canon for the next scene.
Layer 2 — AI for the DM
Reads your campaign first. It already knows Mira suspects Cyrus (a GM-secret you wrote down), so it tells you she wouldn't be surprised by the betrayal, she'd be confirmed, and she'd move on the cult immediately using the evidence she's been gathering. Not a smarter answer. A correctly grounded one.
This is also why "the AI forgot my character's name by session three" is a category problem, not a bug. A tool with no queryable source of truth always drifts as the campaign grows past its context window.
The screenshot section
A solo or async partner that runs the game for you
AI Dungeon, Friends & Fables, AIDungeonMaster.ai, StoryRoll
A campaign manager that keeps your homebrew canon straight across sessions
Grimoire, Multiloop, LoreKeeper, Chronica, Quest Portal
Recorded sessions turned into recaps and entity updates
Archivist AI, Tabletop Arc, GM Assistant, The Enchanted Scribe
Battlemaps from rough ideas
Dungeon Alchemist
NPCs, encounters, loot, and portraits on demand
LitRPG Adventures, RoleForge, ScriptoriumGM, CharGen
A DIY stack you configure yourself
Claude or ChatGPT + Obsidian or Notion
Layer 1
These run the game; you're the player. Genuinely useful if your group dissolved, if you're in a timezone with no table, or if you want to write collaborative fiction with dice. The shared limitation is memory: every one of them hits a ceiling where it forgets earlier events. Don't buy one expecting it to manage a multi-year campaign's canon.
The original of the genre: open-ended text adventures where the AI narrates whatever you do. Great for unstructured solo storytelling, and not really trying to be a rules-faithful 5e referee.
The most D&D-shaped of the bunch. Its Discord bot "Franz" runs 5e-style sessions with leveling, HP, and spellcasting for a party of up to six.
Leans hardest into being a rules-aware engine built around a 5e ruleset.
Focuses on real-time multiplayer AI RPG sessions.
Layer 2
The most crowded layer, and the one Grimoire lives in, so I'll be upfront and detailed. These don't run your game. They hold your campaign's state so you (and optionally your AI) can work with it between sessions. One rule I held myself to: I didn't call Grimoire "the best." I told you what it's for and what it isn't.
More depth: the campaign builder and the RPG campaign manager.
A naming note: there's also a separate tool at thegrimoire.quest that lets you upload campaign documents and ask AI questions about them. Different product, similar name. Its document-upload model doesn't enforce structure the way a typed-entity database does (your AI queries your docs, not a schema), which makes it a lighter setup but a looser guarantee on canon.
Layer 2 · recap
Record the session, get notes back. They mostly do the same job; the real differences are output shape, price, and what they integrate with.
Produces structured notes.
Focuses on evidence-grounded recaps.
Does the audio-to-notes pipeline cleanly.
Trains on your own content over time.
Grimoire doesn't do this natively, so I'll name a recommendation instead of hedging: if you run Grimoire and want recordings turned into notes you can paste back in as entity updates, Archivist AI for structured notes is the cleanest pairing, with Tabletop Arc a close second if you prefer narrative recaps.
Layer 3
These produce assets you copy into your own game. Because they're transactional (pay per generation or by subscription), pricing matters more here than for state tools.
The deep one: a large library of generators for NPCs, encounters, items, dungeons, and more.
Mixes content generation with AI-searchable campaign knowledge.
Generates characters and story content.
The TTRPG-specific one worth the most attention: character portraits, NPCs, monsters, and VTT tokens you can push to Foundry or Roll20, with import from a D&D Beyond link or a PDF. It does what it does well, and it is a genuinely different layer of the workflow than a campaign manager.
General-purpose image generation if you are comfortable steering it yourself.
The open general-purpose option for self-hosted or heavily customized image generation.
Generates usable battlemaps from rough input. Maps are a whole world of their own; this is the obvious starting point.
Layer 2 · manual
The most under-covered option in competitor listicles, and the practical choice for the largest number of readers, so it deserves honesty. A lot of DMs run Claude or ChatGPT directly, with Obsidian or Notion as the database. It works.
The trade-offs: you do the context management by hand every time, the model has no enforced structure so it can contradict your canon when you forget to paste the right note, and continuity is only as good as your prompt discipline. That's the gap a Layer 2 tool closes: structured state, canon enforcement, and cross-session persistence without re-prompting from scratch. The connective tissue is MCP, the open protocol that lets your AI read your live campaign data directly. What the DIY stack gives you that a managed tool doesn't is total control of the data model and offline-first local storage.
Some of you will read this and pick the DIY stack. That's fine; you were probably going to anyway, and being the post that helped you choose clearly is a win even when you're not my customer. If you go this route, write your AI a clear system prompt with your campaign's canon up front so it doesn't drift.
The honest critique
01
The original sin of the category. A list that ranks them against each other is comparing a referee to a filing cabinet.
02
Every AI-as-DM tool, and plenty of AI-for-DM tools, eventually forget what happened earlier in your campaign. The honest answer to "how does this remember a three-year campaign" is "it queries a source of truth, or it does not." Most listicles never ask the question.
03
Most tools have no good answer for what the AI should tell the players versus the DM. If your players can connect their own AI, whether it can reach your GM-secrets is not a nice-to-have. Almost nobody names it.
The decision framework
That last point is the one I'd want you to keep even if you forget everything else: the model market moves every few months, and a tool that bundles a model is stale by design. If a wiki fits your brain better than a typed database, that's legitimate too, and the World Anvil, LegendKeeper, Notion, and Obsidian comparisons lay out where Grimoire does and doesn't fit.
Who wrote this: I'm Zach, the solo developer behind Grimoire. I'm biased about my own product and I've tried to be honest about everyone else's.
How tools were evaluated: I used my own homebrew campaign as the test bed. For tools I haven't run in production, I evaluated based on their documentation, demos, and community consensus. Where I wasn't sure of a claim, I softened it rather than invent a failure mode.
What's missing: I didn't cover Roll20, Foundry, or D&D Beyond, because those aren't AI tools in the sense this guide is about. I didn't cover non-English tools I can't evaluate. And the category moves fast enough that new tools will appear between this writing and your reading.
If I got something wrong about your tool, tell me and I'll update the piece. Last updated June 6, 2026.
Questions, answered
AI as the DM means the AI runs the game: it narrates, voices NPCs, adjudicates rules, and tracks combat while you play. AI for the DM means the AI helps the human GM between sessions (parsing notes, tracking state, drafting recaps, recalling lore) while you still run the table. They solve different problems for different kinds of GM. Most confusion when picking a tool comes from buying one category and expecting it to do the other category's job.
There is no single best tool, because the tools do genuinely different jobs. If you want something to run the game for solo or async play, look at Layer 1 tools like AI Dungeon or Friends & Fables. If you want a campaign manager that keeps your homebrew canon straight across sessions, look at Layer 2 tools like Grimoire, Multiloop, or LoreKeeper. If you want assets faster, look at Layer 3 generators. Pick the layer first; the tool second.
It can run a session-like experience for solo or async play, and some people genuinely enjoy that. What it cannot do reliably yet is remember a multi-year campaign without a source of truth to query, track complex combat state, or adjudicate edge-case rules. Those are the structural limits that separate AI-as-DM tools from AI-for-DM tools.
Often you do not, and usually you should not. Tools that let you bring your own AI client (via the open Model Context Protocol) let you use whatever frontier model you already pay for, with no markup and no per-token quota. Tools that bundle their own AI tend to ship an older mid-tier model marked up to a monthly fee. Bring your own client where you can.
Yes. Several Layer 2 tools have real free tiers, and the foundation-model DIY stack (Claude or ChatGPT plus Obsidian or Notion) is free to start. Grimoire's free tier covers the AI integration and one full campaign forever; you connect your own AI client's free tier on the other end.
If Grimoire sounds like what you're after, the free tier covers the AI integration and one full campaign forever. If you'd rather DIY with Claude or ChatGPT, that's a great choice too. And if you're picking a different tool from this list, I hope this made the choice clearer.