Browser-native AI experiments
Sushi
A client-side AI demo garden where language, speech and visual experiments run directly in the browser.
Independent AI Engineer / Creator · In progress · 2026-Present
View on GitHub
Overview
Sushi is a growing collection of browser-native experiments: local model selection, WebGPU and WebAssembly inference, speech-to-text, text-to-speech and real-time systems that keep the work on the user's device.
Browser lab
Run the experiments
Local-first demos. Models download only when you choose to load them.
- LLMlanguageChoose a local model and chat through WebGPU or the mobile WASM path.Open experiment
- TTSvoiceLoad Pocket-TTS, choose a voice, and synthesize speech in the browser.Open experiment
- LLM + TTSvoice loopConnect local generation to streaming speech and a responsive avatar stage.Open experiment
- STThearingRecord speech and transcribe it with the local streaming or mobile path.Open experiment
- STT + LLM + TTSpipelineRun the complete speak → think → listen pipeline without an API key.Open experiment