Skip to main content

Four steps. Three pointer hops. One answer.

Loom turns the "what failed?" question into a navigation problem, not a reconstruction problem. Instead of scrolling through interleaved output looking for clues, you follow a chain of structured pointers straight to the evidence.

The flow​

1
AuthorYou define your workflow in a single YAML file.
2
ValidateLoom catches schema errors before execution and writes a validation receipt.
3
ExecuteLoom runs your jobs locally and writes structured artifacts at every level.
4
DiagnoseYou follow manifest pointers to the exact failing unit — no scrolling, no guessing.

The pointer chain​

This is the core idea. Every Loom run produces a chain of structured pointers that lead directly from the top-level receipt to the exact failure evidence.

When something fails, you don't search through logs. The receipt points to the pipeline manifest. The manifest points to the failing job. The job manifest points to the exact events file. Three hops from "something broke" to "here's what happened."

What you get back​

ReceiptYour starting point. A small structured file that anchors every diagnostic path.
ManifestsStructured indexes with pointers to exactly what failed and where to look next.
Event streamsThe actual failure evidence. Structured JSONL, not raw log dumps.

Works for you. Works for your agent.​

For you: follow the pointers manually, share receipt paths with collaborators, and reproduce the same diagnosis on any machine. No tribal knowledge required — the artifacts tell the story.

For your agent: the same artifacts are structured JSON with deterministic paths. An AI agent can triage a failure by reading a few kilobytes of structured data instead of parsing megabytes of interleaved text.