How Sensipets Uses AI
Sensipets is made by one person using AI as a creative and technical tool. The useful question is not simply whether AI appears, but where it appears, who directs it, and who is accountable for the result.
The principles I work by
I set the direction
I define the systems, the taste, the safety boundaries, and every final product decision. AI helps make and explore options inside that direction.
Review matches the risk
High-impact work receives direct review. Automated in-world pipelines use constrained inputs, provenance, moderation controls, and operator tools.
Players remain authors of play
AI does not replace the choices that make a pet, profile, collection, shop, forum, or creative brief feel like yours.
Where AI helps
Art and media
AI can help create candidate illustrations, item art, character treatments, music, and sound. I supply the house direction and select or refine what belongs in Veloris.
Writing and worldbuilding
AI can assist with drafts for names, descriptions, dialogue, lore, and news. Constraints and editorial systems keep those drafts grounded in the existing world.
Building the application
AI assists with implementation, testing, review, and investigation. I remain responsible for the architecture, the security boundaries, and the code that ships.
Brightworks makes the process visible
A player begins with an original brief and may choose reference items. Brightworks presents exactly 3 candidate treatments to the player. The player chooses and refines a favorite, and the resulting collectible is reviewed for originality, safety, image quality, and fit with Sensipets before delivery.
Some Communicator replies are written live
When you write your own message to a Communicator contact, AI usually writes the reply as that character, working from a voice and facts I authored. I don’t read these replies before they reach you. I review the ones players report or automated checks flag. Prepared-question answers and the letters contacts send first, such as quest offers and check-ins, are prepared text, not live generation.
Your message gets a safety check first, and some topics receive a fixed response or a warning instead of a generated reply. Each generated reply then passes automated moderation. If it fails, or that check is unavailable, you get a short standard note instead. The model cannot give you items or currency, advance a quest, or change your account, even if a reply sounds as though it did. It can still get things wrong, so use the report flag beside any reply that misses the mark.
To write a reply, the model receives your player ID, recent messages from that thread, and the character’s profile. With Let characters recognize you turned on, it also receives your display name, what that character knows and remembers about you, and your standing with them, and it can save a short note about a preference you state. You can turn this off in Settings → Privacy.
Other text written during play
The Communicator isn’t the only feature that writes text as the game runs. Characters stay in voice and won’t bring it up on their own, so I’m telling you here. Some of this reaches you before I see it, and some waits for my review.
Madame Coraline’s readings
Ask Madame Coraline a question and the game draws three cards at random, then AI writes her reading of them for your question. I don’t read it before you see it. Your question gets a safety check first, and she may decline it or set the cards aside and suggest talking to someone you trust. The finished reading then passes automated moderation. If a check fails or isn’t available, you get no reading and can ask again. The model receives your question and the cards, not your name or player ID, so leave personal details out of your question. Your question and reading are saved to your account so you can reopen them. A reading is for fun, not advice. To report one, use Send feedback below.
Band members at concerts
During a show, AI writes many of the lines band members post in the venue chat, sometimes in response to what players just said. Other lines are prepared. I don’t read the written lines before they post. Each one passes automated moderation first, and a line that fails or can’t be checked doesn’t post. Use the flag beside a line in the chat to report it.
To write a line, the model receives recent venue chat. You appear by username only if Let characters recognize you is on and your Housing identity is shown; otherwise you’re “a listener.” With that setting on, the model can also receive your standing with that character, profile stats you haven’t hidden, your login streak, the shows you’ve attended, and short notes AI makes after each show from what players said in the venue chat. With it off, you stay a stranger and no notes are made about you, but what you say in chat can still shape a band member’s next line.
Guest creators and voters in contests
Ranked contests can include labeled guests from the Sensipets cast. AI writes their entries, draws any art, and scores other eligible entries to form their ballots. I don’t review any of this first, but a guest entry’s title, description, and text must pass automated moderation before the entry can appear. If they’re flagged, that guest sits the round out. Guest art doesn’t get this check. Guest entries are added only while a round has no player entries, and guests can’t win places or prizes. Their ballots count the same as a player’s, so they can change who places. To score an entry, the model sees its title, description, text, and image, not who made it. An entry that tries to instruct the judge is left off guest ballots and logged for my review. Use Report a concern on an entry to flag it.
New dishes, familiars, and crafted equipment
The first time anyone cooks a new combination in the Kitchen, fuses a new familiar, or makes a new design on the Fabrication Deck, AI writes its name, description, and flavor text and draws its art. Game rules decide whether it succeeds and what its stats are, not the model. You pick what goes in, but you don’t type anything to the model, and it doesn’t receive your name or player ID. You may see your discovery before I’ve reviewed it. The text gets no automated moderation, only my review, and it stays out of public listings like the cookbook and Almanac until I approve it.
Workshop projects
At the Inner Cliff workshop, AI writes each project’s name and description and draws its art from the items and direction you chose. Game rules set its rarity and effects. You don’t type anything to the model, and it doesn’t receive your name or player ID. You can’t collect a project until I’ve reviewed and approved it.
Tell me when something misses the mark
AI-assisted work can be inconsistent, generic, or simply wrong. If art, writing, or behavior does not feel right for Sensipets, report it. Specific examples help me review the output and improve the system around it.
Send feedback