Getting Started

SparkLang is a language and runtime for plain .spark files aimed at solo/small teams who want reviewable train→status→expect pipelines in PRs. Start with dry-run (no keys, no network). model train / model build are wired to backend "http" (SPARK_TRAIN_URL): dry-run fixtures first; live needs your trainer. Optional ask against any OpenAI-compatible gateway is not the product identity.

Install the runtime

Download a graphical installer or runtime kit for your platform from Downloads:

curl -fsSL https://sparklang.dev/downloads/spark-install-linux.sh | bash
spark --version

macOS: curl -fsSL https://sparklang.dev/downloads/spark-install-macos.sh | bash. To build from source, download the runtime tarball from Downloads → Build from source, extract it, then run make in that directory.

Run dry-run (no API key)

Dry-run uses offline fixtures — safe for CI and learning without network access:

./spark --dry-run examples/hello.spark

You should see heuristic output (no live model call). Try the Playground page to preview sample output in the browser.

Optional live ask

Only when you want live inference: pass --live and point AI_GATEWAY_URL at any OpenAI-compatible chat API. The spark-ask-http companion does the HTTP call — this does not make Spark a Bifrost install.

export AI_GATEWAY_URL=http://127.0.0.1:4000
./spark --live examples/ask_live.spark

Set the key in your shell or a local env file — never commit secrets. See Run: dry vs live in the programming guide.

Next lesson

Continue to Your First Program to write hello.spark from scratch.