Giving AI assistants a face that reacts: Masko turns mascots into state machines

Paulo posted a Mango demo on X. The orange plush monkey sits in the center of the canvas, smiling with its eyes closed and hair standing up. At the top of the interface is a line of text: 25 poses · 79 clips · one canvas. Below are Hello, High five, Head scratch, plus Nap, Thinking, Hype. Further down are Poke, Drag, Shake, Say a line. In the video, it switches from waving to a high five, then shrinks back to take a nap; the actions are about 2 seconds apart, and the loop is fast.

This system is called Masko. Paulo says all of Mango's animations come from a single prompt, with no rigging. Masko packages it as a state machine engine for AI mascots. Characters in an app don't need to be drawn frame by frame: first generate the character, then generate poses and animations, and then connect them with a state graph: idle, working, celebrating, poked, speaking. It can be used for desktop companions, AI agent faces, and app onboarding.

A quick bit of background. Rive is an interactive animation tool, strong in state machines and runtime control; many products use it for buttons, characters, and loading animations. Masko is targeting another segment: people who don't want to learn Rive and don't want to do rigging, who want to get characters and animations running with prompts. The website puts it directly: upload reference images or describe a character, keep the same character across poses and styles, export transparent video or image sequences, and put them into web, iOS, Android, or AI agents.

On details, Paulo gave a few parameters in replies. Each animation can last from 2 to 30 seconds and can also be sped up. The 2-second pose transition is the default feel, not the only setting. A user asked whether, if a user click happens in the middle of a transition, it blends or jumps directly; Paulo said it can jump, or the animation can continue, and the engine and player are still being completed with the SDK. That answer leaves room. In other words, the state machine engine and player have not yet been fully handed to developers.

Integration is not limited to Studio. Masko offers an API, and the docs include REST, async tasks, Webhooks, and hosted assets. There is also MCP, which Paulo says is the recommended way for agents. On Mac there is a runtime, a CLI, and Swift and TypeScript SDKs, and Node.js and Electron can connect. Export formats include transparent WebM, HEVC MOV, stacked MP4 for Android, PNG sequences, Lottie, and GIF. Public URLs can be dropped directly into an app without hosting them yourself. The Mac app is free; generating images and animations consumes paid credits.

There is controversy too. A user, Salman, said in a reply that Rive looks better and that vector elements have a file-size advantage on mobile. Paulo's response was that you can't easily do everything with Rive. That response isn't forceful, but it hits the difference: Rive gives control, Masko gives generation speed. One requires you to build state machines, tune curves, and manage assets; the other strings character generation, animation, state graph, and runtime into a pipeline. The trade-off is that animation quality has to be reviewed one by one, consistency relies on reference images and a style card, and the blending strategy for complex interactions has not been made public.

Another user asked about voice. Paulo replied, 'Of course there is voice.' Asked about 3D and API, he said there is no 3D, and MCP is available. Asked whether the animations were designed with Rive, he said Masko's animations are not designed with Rive. Put these replies together, and Masko's boundaries are fairly clear: 2D characters, transparent assets, state-driven, aimed at apps and agents. It does not intend to replace professional animation software, at least not now.

If you've built empty states, loading screens, or onboarding pages in an app, you know a little character that reacts can absorb a lot of awkwardness. The problem is that most teams don't have an animator and don't have time to maintain a state machine. Masko lowers the barrier to a prompt and a few reference images. What remains to be seen is whether, after the state machine engine ships, interruption, blending, and performance can hold up in real products. Rive users will be watching this closely.