The game is designed with a highly modular architecture, where each major functional component can be replaced or upgraded independently. The core components include Whisper system for speech recognition, Kokoro TTS service for speech synthesis, and gpt-oss language model. Developers are free to use locally deployed models or connect to cloud APIs according to their own hardware conditions and needs.
For local deployment, the project supports running the gpt-oss-20b/120b model via llama.cpp, but requires the appropriate hardware configuration. For low-performance devices, the .jinja configuration file can be adjusted to reduce the inference level; for users seeking high performance, the cloud API can be replaced or the inference level can be modified. This flexibility makes the project suitable for both privacy-sensitive individual users and development needs that require higher performance.
This answer comes from the articlegpt-oss-space-game: a local voice-interactive space game built using open-source AI modelsThe
English 
简体中文 
日本語 
Deutsch 
Português do Brasil