Why it’s interesting
Wovi Agent Maze is not an open-ended child chatbot. It turns a small reward-scoring algorithm into a playable maze. Wovi must reach a sky beacon, but the fastest route contains both energy and a storm. You predict whether it will take the safe route or chase the shortcut, then run the starter policy. In our no-login test, the agent hit the storm after two steps: the page showed that +2 energy cancelled the −2 storm penalty, leaving a +4 tie that the fixed scan order broke toward danger.
After the failure, you can raise the safety penalty from −2 to −12 and enable a reward for unvisited tiles before rerunning. Every visual control maps directly to the displayed Python class, variables, conditions and score function. The lesson is therefore inspectable: goals are designed, rewards can conflict, behaviour is evidence, and code makes the rule precise.
How to try it
Open Agent Maze without registering. Make a route prediction, run the policy and inspect the progress, energy, safety and memory scores. Raise the storm penalty after the starter policy fails, then run the revision. The mission takes about five minutes and is designed for ages 9–13 with an adult, but also works as a first reinforcement-learning mental model for adults.
Requirements and sources
The current playable prototype is free and requires no account. Wovi says progress stays local, analytics use a random session identifier and cookieless events, and it does not collect children’s names, emails or birthdays. The separate waitlist is for a parent email and is not needed to play. The site labels itself a pre-launch concept; accounts and persistent progression remain future plans, so this listing does not present it as a finished commercial course.
[Play Agent Maze](https://wovi.world/agent-maze) · [Current status for parents](https://wovi.world/for-parents) · [Privacy](https://wovi.world/privacy)