Skip to content
This site is under heavy WIP, so contributions on GitHub are much appreciated! You've most likely been pointed to this site to point towards a concept, or something. Either way, take at least some of the info on this page with a grain of salt, and also don't expect much info since it's very incomplete on content.

API Overview

Official Neuro Game SDK icon

Official Neuro Game SDK icon. Source: VedalAI/neuro-game-sdk. These docs are not official.

The Neuro Game API, despite its name, isn’t an actual API; rather, it is a protocol on how to communicate game states, actions and more to the Neuro twins. Specs were open-sourced to allow for fan-made games to be playable by her (a prominent example is some games made during the Neuro-sama Game Jam 2 post-jam.)

If you’re familiar with LLM tool calling, the Neuro API is similar to that, so the concepts of tool calling will more or less apply here.

Neuro herself doesn’t directly interact with the game, instead, a WebSocket server acts as middleware/a relay between Neuro and the game, allowing commands to be sent back and forth via the server. Despite this, however, there are some limitations:

  • Neuro cannot understand messages in binary format. This limitation is not present in Randy and may not be present in other Neuro simulators.
  • It is unknown whether or not Neuro can accept multi-connects without issues (the current assumption is that she can)
  • The API is optimized for turn-based games. It is not recommended to try this API with real-time games.
  • You need to be able to describe the entire game state in text to Neuro. You shouldn’t rely on Neuro’s vision module being on while playing your game.
  • Quoting Alex himself, “Vedal said you can use this for more complex games but he told me “you wouldn’t get it” so I’m not even going to try to imagine how this would even work, and I am just going to assume that he is wrong.”
  • alexvoid (VedalAI/neuro-game-sdk), quoted text can be found here (expand the Examples dropdown).