Playground

Edit .spark source, preview dry-run output in the browser, and try a browser voice demo (Web Speech API). This is an honest preview — not a cloud IDE. Live model and voice ops run on your machine with the Spark runtime.

Your .spark program

Output console (simulated dry-run)

Run dry-run to see simulated fixture output for ask, classify, model, and voice ops.

Live (local only):

Download or build the Spark runtime
Save your program as my_program.spark
Dry-run: ./spark-run.sh --dry-run my_program.spark
Live: export OPENAI_API_KEY=your_key then ./spark-run.sh --live my_program.spark

API keys stay on your machine — never paste them into this page.

Pick a default model alias. Changing the alias updates the use line in the Code editor and shows what model analyze, compare, improve, and build would return in dry-run.

Default model alias:

Dry-run preview

Full workflow snippet

Browser voice demo — uses Web Speech API where supported. Spark listen / speak ops run locally with ./spark-run.sh --live.

Transcript

Assistant reply

Example voice program

listen -> user_text
ask "Reply briefly: {user_text}" -> reply
say reply -> "out.wav"

Playground

Dry-run preview

Full workflow snippet

Example voice program

Next steps

Model builder wizard

Getting started