Each capability is one job your app can hand to a worker, and it returns a structured result your app can use. One has a working demonstration of how installing works — it has not been measured and we do not sell it as an agent. The rest are honestly marked as coming next — no fake buttons.
“Can an AI add points, levels, streaks, quests and badges without making my app childish?”
Describe what users do in your app. Get the game loop, rules, events, data shape and UI-ready copy.
Grouped by the job each one does. None of these is installable yet, and none of them is measured.
“Can an AI turn my app into an adaptive challenge or riddle game?”
Difficulty curves and puzzle sets that adjust to how a user is actually doing.
“Can an AI add safe characters that react to each user?”
A cast with fixed boundaries, so a friendly guide never drifts somewhere it shouldn’t.
“Can an AI design onboarding that changes based on what a user does?”
A first-run path that branches on real signals instead of a fixed five-step tour.
“Can an AI score user-created work consistently?”
A rubric that gives the same answer twice, with the reasoning attached.
“Can an AI check my app’s copy and rules for harm before users see them?”
A review pass over the mechanics and words your users will actually meet.
“Can an AI turn the feedback my users send into something I can act on?”
Turn your feedback button into a weekly list of what to fix
Audio is one governed family, not a pile of separate experiments. Every part of an audio job is fixed by the installation and owned by a server: which character speaks, which narrator reads, which cue keys exist, what a microphone may be used for, how long an asset is kept and what rights it was cleared under. An app never picks a voice, never sends words for a machine to compose against, and never holds a credential. They are grouped by wave below because they arrive in that order, not all at once.
The bounded jobs whose input is already owned by a server: an approved reply, a fixed cue key, a saved summary.
“Can my app speak an approved reply in a consistent character voice?”
Renders a persisted character reply as private audio. The character is fixed by the installation: no arbitrary voice selection, no voice cloning, and no provider to name in your app.
“Can my app play a consistent sound when something happens in the game?”
A fixed cue pack your app can play back: success, near miss, unlock, warning and reveal. The cue keys and the way each one is made are owned by the server, not by a prompt from the browser.
“Can my app read back the summary instead of making someone read it?”
Turns a summary or decision brief your app already saved into a private spoken recap, with one fixed narrator.
Each one needs wave 1 first: an authorized capture, an asynchronous render, or an existing render to align against.
“Can someone speak their answer instead of typing it?”
Short microphone input becomes text your existing gameplay already understands. The audio is not kept by default, and your game rules stay in charge of what the text means.
“Can my app turn an approved script into a scene with more than one speaker?”
An approved, bounded multi-speaker script becomes a cinematic dialogue render. It is produced asynchronously and it is not an open-ended live agent.
“Can my app highlight the words as they are spoken?”
Aligns an approved script with the audio your app already rendered and returns the timing map behind word highlighting and captions. Timings only: nothing about how the audio was made comes back.
Each carries an open rights, cost or safety question, and none of them ships without its own written gate.
“Can my app have a short music bed that matches the mission?”
A short, server-owned music bed generated from a fixed mood key and cached as one asset. The browser never sends words to compose against, and nothing starts playing on its own.
“Can my app reach people who do not read English?”
Translates an approved audio or video asset into a chosen set of supported languages. The job runs asynchronously and the installing app attests, per asset, that it holds the rights.
“Can an authorized adult perform a character instead of typing it?”
An authorized adult creator records their own delivery and it is transformed into a rights-cleared designed character. There is no arbitrary target voice and this is never offered to learners.
“Can my app hold a spoken back-and-forth that can be interrupted?”
Experimental. A governed spoken agent with turn taking and interruption, wrapped around a model you bring. It is bounded by a purpose and a time limit, and it is explicitly not a persistent companion for a child.
Direction, not delivery. The catalog above is the only statement of what exists.
Coming-next capabilities have no sign-up and no waitlist yet. They are here so you can see the direction, not to collect your email.
Our published reference check covers an earlier agent — banking request routing — and shows the method: trial tasks, a declared standard, and the misses printed next to the passes. Gamification Builder’s own check is not published yet.