Build a Model

Use Spark to analyze what a model does today, compare candidates on your eval suite, improve toward quality / speed / cost / local, and build a blueprint file you review before training. This wizard generates a copy-paste .spark program for either path.

1. 1. Choose path
2. 2. Set options
3. 3. Copy .spark snippet
4. 4. Run locally

Step 1 — Modify existing or create new?

Step 2 — Options

Step 3 — Generated program

Copy into my-model.spark:

Step 4 — Run

Dry-run uses fixtures — no API key, no training jobs. model build writes a plan and config markdown file, not model weights. Live model calls need OPENAI_API_KEY and ./spark --live.

Full model workflow reference

model analyze "fast" -> report
model analyze all -> catalog_report

model compare ["fast", "code", "best"]
  on suite "examples/eval_suite.json" -> comparison

model improve from report prefer quality -> blueprint
# prefer: quality | speed | cost | local

model build blueprint into "out/better-model.md"

Sugar shortcuts: use fast, ? "…" for ask, and spark.toml for a default model. Full spec: Language reference § model.

Preview in browser

Playground includes a Model tune workflow example with simulated dry-run output for the full analyze → compare → improve → build chain.