Boundless AI Fantasy Review

A Dungeon Master Made Of Language Models

Boundless AI Fantasy released on Steam on September 11, 2026 from solo Brazilian developer João Vitor Andrioli de Souza, and its pitch is unusually specific for an indie RPG: it's a front-end for large-language-model-driven tabletop-style roleplaying, wrapped around a world-and-campaign editor and a supporting layer of isometric, grid-based turn-based combat. Players write their own settings, characters, and backstories, hand that material to a language model as narrative context, and then play through the resulting story, dropping into tactical combat whenever a fight breaks out. This is a feature the developer advertises directly on the store page rather than something buried in fine print, and the review pool engages with it just as directly. At $9.99, it launched to a small English pool of 9 reviews, 7 positive and 2 negative, all 9 read in full for this review.

The Editor Is The Real Product, And Reviewers Who Get That Tend To Like It

The reviewers who respond best to Boundless AI Fantasy are the ones who treat the campaign editor, not the combat, as the actual game. "This game is something unique. I've been writing stories for months, having backlog and backbones for campaigns, and actually getting to play some of them," wrote one reviewer, calling the result "simply wondorous." Another described feeding the game's world-map editor with lore for reuse across settings: "easy customizable in every franchise you want. Feed it with enough lore and go." A third specifically praised the flexibility to bring outside language models into the mix, noting the game includes a menu button that links directly to Hugging Face for adding your own model rather than locking players into whatever ships by default. Combat, in this framing, is explicitly secondary: one detailed review described it as "simple but serviceable... the purpose of the game isn't the combat itself, that was added to help flesh out the roleplay, which is the main draw."

Two Very Different Audiences, Two Very Different Verdicts On The AI Itself

The clearest fault line in this pool isn't positive versus negative, it's novice versus experienced when it comes to LLM-driven roleplay. One reviewer, self-described as familiar with the wider world of AI-driven roleplay tools, praised the game as approachable "for someone who is a novice to LLM-driven roleplay" but flagged a real limitation for advanced users: the game offers no direct tuning of core LLM parameters like temperature, top-k, or repetition penalty, leaving output quality entirely dependent on which model a player chooses and can't adjust from inside the game itself. That same reviewer tested several models and reported one performing poorly, producing rich prose but repeating lines verbatim and stalling the narrative. A second reviewer made a related but more basic point about the free, built-in smaller models specifically: "the smaller models... are very limited and pretty much just repeat back to you what you tell it. If you say there is a mysterious crystal, the game will agree that it is a mysterious crystal and never push the story forward," concluding that the experience depends heavily on the player being self-motivated to drive the plot themselves rather than expecting the AI to do it for them. The pool's two negative reviews land exactly where this thread predicts they would: "basically worse than just using a regular web browser based AI," and, more bluntly, "I love the concept. the AI is very stupid. Makes it worthless."

A Real, Reported Technical Limit On How Much Lore It Can Actually Hold

Beyond model quality, one reviewer reported a concrete technical ceiling worth flagging directly: attempting to import or describe even a single detailed city's worth of lore into the World Map editor caused the game to crash, unable to handle the length of text. That same reviewer still recommended the game and called it "good for the price," framing the crash as evidence of a promising but currently limited tool rather than a broken one, and specifically hoped for future support for "stitching thousands of pages of lore together for different cities, nations, peoples, and more." For a game whose entire premise depends on players building rich, detailed settings for an AI to narrate from, a text-length ceiling this low is a real constraint on exactly the audience the game is trying to attract, players with existing campaign notes and worldbuilding they want to bring in wholesale rather than write from scratch inside the tool.

Verdict

Boundless AI Fantasy is a genuinely novel pitch, an LLM-driven Dungeon Master wrapped in a lightweight tactical-combat layer, executed by a single developer with real strengths and real, specific limitations that this small review pool documents clearly rather than vaguely. It's a strong fit for newcomers to AI-driven roleplay who want a structured on-ramp with pre-built characters and campaigns to start from, and a frustrating one for experienced users who already know what they're missing without in-game parameter tuning and want to bring in existing worldbuilding that the editor currently can't handle at scale. At $9.99, with the option to plug in outside models via Hugging Face for players willing to go looking, this is worth a look for anyone curious about AI-assisted tabletop-style storytelling on Steam specifically, provided expectations are set by the model you bring to it rather than the ones it ships with. One reviewer summed up the split verdict about as fairly as this small pool allows: it's less a finished game than "a gateway" into a much larger, still-emerging category of AI-driven interactive fiction, worth the asking price for a taste of that even if the destination it points toward currently lives mostly outside Steam altogether.