aiga_ Review: The First AI RPG You Can Lose to a Stranger
Every AI RPG we have scored is a story you tell yourself. aiga_ is the first one you can lose to somebody else. Its Kingpin mode gives a compatible world a leaderboard, gives you a reputation score on it, and ranks that score against every other player in that world and against the site as a whole. Your job is to build as much rep as you can and not die. That is a competitive game, played inside a generated story, and nothing else in this directory has tried it.
It scores 2.9 out of 5, eleventh of the twenty-five platforms on our benchmark, level with MacerAI. The gap between that number and the paragraph above it is the whole review — and it is a different gap from the one we usually report. aiga_ is not a thin design dressed in ambition. It is a good design in a build that is not finished, and it is the first entry on this board whose composite is set mainly by bugs rather than by decisions.
What It Is
aiga_ — from AI WORX LTD, and originally “AI Game Agent” — is a browser platform where you either play a world somebody else published or write your own from a short prompt. It runs in sixteen interface languages, keeps a per-world record it calls the Living Game Book, and will hold your character’s face consistent across scenes and across a dozen art styles from a photo, a sketch or, in its own example, a toy you photographed.
Each world runs in one of four shapes, and this is where it separates itself from the field. Four modes, described here as the platform describes them:
| Mode | What it is |
|---|---|
| Interactive Story | A guided, chapter-like narrative where choices shape what happens next |
| Open World / RPG | Freeform exploration with characters, locations, factions, objectives, inventory and evolving world state |
| Single Objective | A focused scenario built around completing one clear goal |
| Kingpin | A competitive mode a compatible world can run: players build power and reputation and are ranked against other runs |
Most platforms in this category ship one of those and call the others a roadmap. Shipping all four is the first sign of what this entry keeps demonstrating: somebody here had a lot of ideas and got most of them built.
The same is true of where you can play it. Solo is private and self-paced. Private groups bring in friends and family with configurable voting on shared choices. Community games are open worlds anybody can join and vote in. And then it leaves the website entirely: a Discord server can run a game with interactive story voting and live updates, a timed poll on X can hand the next decision to your followers, and a Telegram chat can manage a game with the voting handled on the web. For a category that mostly means “one person and a text box,” that is a genuinely different product.
Signature Design — 4: The Leaderboard Is Real
This is the axis that asks whether a product does something nobody else does, and Kingpin answers it cleanly.
One thing to be exact about, because it is easy to get wrong: Kingpin is a mode, not a world. It is not a separate competitive place you go to — it is a setting a compatible world runs, and the worlds that run it get a leaderboard. Everything below describes what turning it on does.
The important part is not that there is a score. It is that the score is authoritative and ranked. Each run reports its own progression score, the world keeps the best and current ranked entries, and seasonal views can reset the competitive window without deleting the underlying game — so a season ending does not take your campaign with it. Leaderboards attach to the individual worlds running the mode rather than collapsing into one global number, which is the right call: a reputation earned in a cyberpunk syndicate world is not commensurable with one earned in a fantasy court, and pretending otherwise would have made the whole thing decorative.
aiga_ also keeps the distinction that most platforms would have blurred. Ordinary story and RPG worlds still display popularity and play counts — but those are labelled discovery statistics, not a competitive board. A platform that wanted engagement metrics to look like achievement would have merged the two. This one did not.
What keeps it at 4 rather than 5 is that the competition sits on top of the game rather than inside it. Reputation is a number the world hands you for doing well in it; it is not a resource other players can take, contest or spend against you. Kingpin is a scoreboard around a single-player run, not a shared world where your rivals are actually in the room. That is still more than anyone else here has built, and it is one design decision away from being much more.
Around Kingpin sit the other things that earn this score: the image continuity, which solves a problem most illustrated platforms simply live with; the sixteen languages, which is infrastructure rather than a feature; and the four-mode structure itself.
Determinism & Fairness — 2: The Bug in the Ledger
Here is the score that decides the composite, and the reason it is worth writing this review carefully.
The normal failure on this axis is a narrator that agrees with you. A language model produces the most plausible continuation, refusal is rare in training data and reads as unhelpful, and most products have no state underneath from which a refusal could be justified — so the guard believes you, the lock opens, and the wound turns out to be survivable. Nineteen of the thirty-seven products we have scored sit at exactly 2 on that axis, and nearly every one of them is there for some version of that reason.
aiga_ sits at 2 for a different one. The bugs reach the ledger. Rewards and purchases can come out wrong. It would otherwise be a 3 — the honest neutral score, where outcomes are neither notably earned nor notably arbitrary — and it is not, because the arithmetic between what you did and what you ended up holding is not yet reliable.
That distinction matters for what you should expect from it, and it does not matter at all for how it feels. From the player’s chair, a reward that does not arrive and a guard who believes an implausible lie produce the same thing: a session where what you did stopped determining what you got. The axis measures the outcome, not the cause, which is why this scores where the generous narrators score despite being nothing like them.
It matters enormously for what happens next, though. A platform that is unfair because it has nothing underneath from which to say no has an architectural problem, and architectural problems in this category have taken years. A platform that is unfair because a transaction miscredits has a defect list. Those close.
There is a second fairness question here that is not a bug at all, and it belongs in the open. In community games everyone can vote, and each turn counts the first N votes it receives — where N is set by your subscription tier: five on the free plan, ten on Explorer and Bronze, twenty on Silver, fifty on Gold. As a way of pricing server load that is defensible. As a mechanic it means a paying player carries more weight in a shared story than a free one, and anybody organising a community around this should know that before they start rather than after somebody notices.
NPC Fidelity — 2: Characters Trading Lines
The weakest axis, and the one where the bug is most visible in the fiction rather than in the numbers.
Characters here do not have a very distinct voice. That alone is an ordinary mid-field complaint — casts flattening into a single narrator wearing several names is the most common failure on this axis across everything we have measured, and it is what keeps most products at 2 or 3.
What is not ordinary is the second half: a bug sometimes has one character speak another character’s lines. Not converge on a shared register — actually deliver dialogue belonging to somebody else. It is the same failure the whole field has, dropped one level down the stack: instead of a model that cannot keep two personalities apart, an implementation that cannot keep two speakers apart.
The consolation is the same as on fairness, and it is a real one. Flattening is hard because it is a property of how these models write. Misattributed dialogue is hard the way a hard ticket is hard. We would rather review the second problem.
Mechanical Depth — 3: Good When It Works
Mechanical Depth — 3 is above the field, not below it. Twenty-four of the thirty-seven products we have scored sit at 2 or lower on this axis, which is the worst average of our seven by a clear margin, and it is worth saying plainly that aiga_ clears it.
The systems are present and they are thought through. Objectives, inventory, factions, locations and an evolving world state are all really there in the Open World mode, and they behave like systems rather than like nouns the narrator mentions. When they work as they are supposed to, they are really good — and that clause is doing the work in this entire review, because the reason this is a 3 and not higher is that the qualifier is needed at all. A rules layer you cannot fully rely on is worth less than the same layer you can, and the score reflects the version we played rather than the version described.
Player Agency — 3 is mid-table for the ordinary reasons. You are mostly free to do what you want. The game still tries to send you off toward the quest it would prefer you were doing — the softer half of railroading, a nudge rather than a wall. And there are one or two remnants of puppeting left in it, where the narration takes a turn your character did not decide to take. Neither is as bad as the worst of the field, and neither is absent.
Memory — 3 and Longevity — 3
Memory & Continuity — 3 is regular for a world that is a prompt plus an engine. The Living Game Book keeps each world’s own history — alliances, character relationships, your hero’s journey — and in play it neither surprised us nor let us down. It is not the state-outside-the-model architecture that the platforms at the top of this axis use, and it does not pretend to be.
Longevity — 3 is the axis where Kingpin should have paid off most, and it half did. A ranked score is the strongest reason this category has yet produced to open a game on day thirty, and the underlying game is solid enough to support the ambition. What holds it at 3 is the bugs again, and specifically what they do to motivation: a run that mechanically miscredits you, or that breaks immersion by putting the innkeeper’s line in the guard’s mouth, is a run you are less likely to resume. In a competitive mode that is worse than it sounds, because the thing you are being asked to invest in is a standing you can lose by not playing.
Pricing
The pricing has one genuinely good idea in it, and it is the one most likely to be missed: text play is included with every paid plan. Creating worlds, starting games, taking story turns and sending chat messages cost no credits at all on a subscription. Credits exist for generated images, video clips and other media-heavy extras. In a category where the standard shape is a meter on the thing you actually came to do, that is a real difference. Verified on aiga.io, 8 September 2026:
| Plan | Price | Media credits/month | Roughly | Votes counted per turn |
|---|---|---|---|---|
| Free | $0 | 0 | Text only, 5 turns/day | 5 |
| Explorer | $8.99/mo | 0 | Text play, no media | 10 |
| Bronze | $11.99/mo | 3,000 | ~30 images or ~7 video clips | 10 |
| Silver | $23.99/mo | 14,000 | ~140 images or ~35 video clips | 20 |
| Gold | $59.99/mo | 50,000 | ~500 images or ~125 video clips | 50 |
The free tier is five text-only turns a day with no card required, which is enough to find out whether the writing suits you and not enough to test a campaign. Explorer at $8.99 is the tier the pricing page half-hides and the one most players will actually want: unmetered text play, no media allowance, nothing else. If you never generate an image, it is the entire product for nine dollars.
Extra credits can be bought at any time. Plans change at any time — upgrades prorated against the remaining cycle, downgrades from the next one — and cancelling keeps access to the end of the paid period. Refunds are handled case by case by support. There is an enterprise tier for large-scale community voting, quoted on contact.
No underlying model or provider is named anywhere on the site, which is worth knowing if model transparency matters to you.
There is no app on either store. The site ships a web app manifest with standalone display and maskable icons at 192 and 512 pixels, so it installs to a home screen and opens without browser chrome — the same route ten other platforms in our directory take.
Who It Is For
Play it if you want a reason to come back that is not just “what happens next.” Kingpin is the only competitive answer anybody in this category has shipped, and if a leaderboard is what makes a game stick for you, nothing else here offers one. Play it too if you run a Discord server or have an audience on X — the voting integrations are the real thing, and this is the least awkward version of group AI roleplay we have used that does not require everyone to be in the same tab.
Skip it if you need the numbers to be right. If you are the kind of player who tracks what you earned and notices when it does not arrive, the current build will annoy you specifically and repeatedly. Skip it too if distinct NPC voices are what you come for — Janitor AI, Tidefall and Voyage all score 5 on that axis and this one scores 2.
Come back to it in a quarter if the ideas appeal and the polish does not. That is not a dodge; it is what we plan to do. The defects here are on the list that shipping closes, and the design underneath them is better than the score.
Verdict
aiga_ has more ideas than most of this board and less polish than nearly all of it. Four game modes that all exist, a competitive layer with an authoritative ranking, cross-platform play that turns a Discord server into a table, image continuity, sixteen languages, and a subscription that does not charge you to take a turn — that is a lot of product, built by people who clearly thought about the problems. What has not happened yet is the unglamorous part: making the rewards land every time and keeping each character’s lines in their own mouth.
We score what shipped, so it scores 2.9. But this is the first entry we have published where the number is set by a defect list rather than by a design, and those two things have very different futures. 2.9 out of 5 — and a re-score expected next quarter.
Full axis scores, the mode table and the pricing detail are on the aiga_ directory entry. How we arrive at these numbers is on our methodology page, and the benchmark reports what all twenty-five platforms look like together.
aiga_ granted us subscription credits to play the platform. Arcanum is independent and unpaid; no commercial relationship exists between aiga_ and Arcanum, and no conditions were attached to this coverage. Product names and trademarks are the property of their respective owners. Arcanum is not affiliated with aiga_.