What is Kiwikidoo? Screen-free interactive voice adventures for children

Kiwikidoo™ is the interactive voice adventure platform developed by yeshcube for children. A bounded conversational artificial intelligence engine builds stories that move forward with each child’s spoken decisions, without putting a screen at the center of the experience. The product is designed with the Somia™ Within framework, which applies common criteria of conversational safety, privacy and evidence. The final application is in development and its validation is progressing through the levels of the hub’s ERL scale.

The ERL scale (Evidence Readiness Level) measures the evidence maturity of a solution across six levels, from the initial hypothesis (ERL-0) to consolidated impact (ERL-5).

The problem: passive digital entertainment for an age that calls for participation

A good deal of children’s digital entertainment is consumed by watching: video on demand, screen games and platforms of continuous recommendation. Many families are looking for alternatives that reduce that visual exposure and give the child an active role, with room to talk afterward about what happened. The project’s earlier research, gathered in the report on its first public test, documents that interest: parents asking for tools to talk about emotions and values without putting a screen in front of them.

Audio solves part of the problem. An audio story removes the screen, but fixes a single path: the child listens to a closed narration in which their decisions play no part. At the other extreme, open conversation with a general-purpose model is ruled out on safety grounds, because a system with no limits of theme, age or supervision is unsuitable for children. Kiwikidoo occupies the space between the two: participatory narrative with defined limits.

Design requirements

Five requirements follow from that problem, and they govern Kiwikidoo’s design:

• Voice as the only interface: the experience is heard and spoken, without constant visual support.
• Bounded narrative: each adventure operates within themes, characters and limits appropriate to the configured age.
• Adult frame: the account, the age, the permitted themes and the restrictions on use are always defined by the responsible adult.
• Conversational safety: every generated response passes a filter before it joins the story.
• Minimal data: only what is strictly necessary is collected about the child, and only for as long as it is needed.

Architecture: the narrative engine and the Somia Within framework

Somia’s modular character separates what the core provides from what each product resolves. In that division, Kiwikidoo provides the stories, the physical device and the family application, while the Somia Within framework carries the conversational safety rules, privacy criteria and evidence requirements defined by yeshcube over to the product.

The artificial intelligence operates as a specialized narrative engine: it develops each adventure from a starting point, a theme and defined objectives, and the child’s decisions change the course within a structure delimited in advance. Faced with sensitive content or anything outside the story, the system gently steers the conversation back and offers a safe, age-appropriate ending.

The physical support for the experience is a pair of headphones or a soft toy with the equipment built in. A light and sound indicator shows when the system is listening, and a physical mute button cuts the microphone without going through the app.

How an interactive voice adventure works

A session follows a defined path. The adult creates the account, states the age and configures the permitted themes and the restrictions on use. The child hears the opening situation and answers in their own voice: asking, deciding, getting things wrong and trying different paths while the story takes their contributions on board. Every generated response passes the safety filter before it is spoken, and delicate themes such as fears, pain or loss are closed safely within the story itself.

At the end, the family application receives a structured summary of the adventure: the main moments, the decisions taken and questions to talk about afterward. That material turns each story into a starting point for conversation at home or in the classroom, with interpretation and support always in adult hands.

Data and privacy

Kiwikidoo’s data regime starts from a written commitment: children’s voices are not used to train artificial intelligence models, neither our own nor third parties’. The raw audio of each contribution is processed during the adventure and discarded at the end: by default it is not kept. What does remain is the structured summary the family sees.

The system stays out of psychological assessment: it does not label, does not diagnose and does not generate clinical markers from the child’s voice or decisions. No children’s data is used for targeted advertising. Full details of the processing, including processing on servers inside the European Economic Area wherever technically possible, are published on Kiwikidoo’s safety and responsible AI page.

Family and educational use

The adventures work for individual use, family listening and group activities. At home, the experience fits into specific moments: a bedtime routine in which the child lives the story with their headphones and afterward talks with a parent about what they decided and why. In the classroom, teachers can set the same adventure as a shared activity and work on the decisions taken: living together, oral expression, or analyzing a situation from several perspectives.

In both contexts the technology organizes the narrative interaction. Interpretation, dialogue and the connection with real experience remain the responsibility of the adults accompanying the activity.

Evidence and validation status

Kiwikidoo was developed within UNICEF Lab, the social innovation accelerator of UNICEF Spain, where the product definition and its fit with the needs of children and families were worked out.

The first public evaluation of the format took place on 9 May 2026 at Expociencia, at the Parc Científic of the Universitat de València, with an adventure tested by children aged 4 to 12. It was an exploratory test aimed at observing the initial reception of the format, with a small sample and fair-ground conditions.

Kiwikidoo holds level ERL-1, proof of concept: a prototype assessed in a first exploratory test with children and families, with initial metrics gathered in a real setting. Level ERL-2 (controlled pilot) requires pre-registered protocols, validated metrics and formal ethical approval in bounded contexts. It will be declared once that evidence is available. Commercial transfer is subject to the hub’s general rule, which requires at least ERL-3 to transfer any solution: the final application, currently in development, is going through that validation process.

Kiwikidoo applies the Somia system’s criteria to children: bounded stories, participation by voice, adult supervision and declared evidence at every stage. The architecture behind it is described in the Somia article, and the product details are on the Kiwikidoo site.