open ./06_ollama.html

Same prompt, two backends. Mix Ollama API models and local WebGPU models side by side — compare quality and speed across different backends and sizes.

setup configure backends // mix ollama + local as needed

Ollama URL

Not connected

If connection fails, restart Ollama with CORS enabled: OLLAMA_ORIGINS='*' ollama serve

Model A

Select .task file

Model B

Select .task file

Load whichever .task files you have selected — requires WebGPU (Chrome 113+).

prompt // ctrl+enter to generate

Model A

✓

—

tokens: 0 time: 0.0s speed: —

Output will appear here…

Model B

✓

—

tokens: 0 time: 0.0s speed: —

Output will appear here…