Skip to main content

BlazerOps

BlazerOps runs pipelines — directed graphs of typed nodes that read your organisation's data, run code, and write results, with lineage recorded for every artifact a run produces.

You work with BlazerOps two ways:

  • The CLI (blazerops) — author a pipeline as a pipeline.toml file, compile it, validate it against the server, push it as a version, and run it. This is the path this documentation covers first.
  • The canvas — author and run the same pipelines visually in the BlazerOps web app.

Both produce the same thing: a versioned pipeline that runs on the platform.

Core ideas​

A few nouns recur throughout these docs:

  • Pipeline — a named, versioned graph. Each push creates a new version in the pipeline's lineage.
  • Node — one step in the graph. A node declares its inputs, outputs, and where it runs; it does not orchestrate other nodes.
  • Kernel — the code a run_code node runs (process(item, ctx)), tracked as a named lineage of versions.
  • Datawell — your organisation's data plane. Nodes read and write it by artifact id; the platform resolves ids to storage and mounts them, so your code sees local files, never storage URLs or credentials.
  • Model — trained weights, registered so a node can load them by pinning a version.

Next​