January 6, 2026
Snake eats hype, comments bite back
My Tamagotchi is an RL agent playing Slither.io
Digital pet plays online snake; fans cry 'no Tamagotchi!' and nerds call the bot buggy
TLDR: A hobbyist used AI help to build a bot that plays Slither.io by steering the browser like a mouse. Commenters split between Tamagotchi nostalgia and disappointment, skepticism about buggy reinforcement learning, and calls to scale training with many tabs—raising fun questions about cheat bots and AI-assisted coding.
The post promises a digital pet mastering Slither.io, built with help from AI tools (Cursor and Claude), steered by simple browser tricks to move the mouse and read the game’s state. It even runs on a Raspberry Pi, with a monitor blurred by Gemini. The dev admits it’s “vibe code” and leans on AI assistants, raising eyebrows.
But the comments stole the show. Nostalgia exploded: giancarlostoro detoured into Tamagotchi-vs-Digimon 90s lore, while remix2000 cried a classic bait-and-switch—“Clicked for Tamagotchi, but I saw none”—the crowd felt click-tricked and amused. On the tech side, jeremysalwen went full skeptic, warning reinforcement learning (trial-and-error training) is notorious for hidden mistakes that just make agents bad, not broken. breakds chimed in with the “it only gets good with scale” take, pushing for many tabs to train faster. treyd cheered, saying they had the idea in lockdown and love seeing someone ship.
The subtext: Is a bot in a public game cheating or just playful science? Does AI-assisted coding count as “doing it yourself”? The community turned this snake project into a showdown of nostalgia, expectations, and buggy RL realism—with jokes, side quests, and plenty of “show me the Tamagotchi” energy.
Key Points
- •An RL agent was built to play Slither.io by automating the browser and injecting JavaScript for control.
- •Selenium loads the game, and JS sets window.xm/window.ym plus mousemove events to emulate mouse direction; boost click is not used.
- •Observation data is read from global JS variables (window.snake, window.slithers, window.foods, window.preys) to avoid computer vision.
- •Actions are discretized to simplify the agent’s control space, and engineered feature vectors are used as inputs.
- •The project leveraged AI tools (Cursor, Claude) for development and documentation, runs on a Raspberry Pi, and references OpenAI Gym/Universe as context.