Before: Log archaeology
Noise-heavy… 5,000 lines of interleaved stdout …
error: command failed (exit code 1)
… which job? which step? search and hope.Interleaved text output forces you to reconstruct the failure path after the run is already over.
ALPHA 1
Run workflows with pointer-first diagnostics.
Loom is an open-source workflow engine that writes deterministic receipts, manifests, and event streams so you — or your agent — can jump to the exact failure without digging through scattered logs.
3 pointer hops
from run to failing step
~2 KB
to the failing unit
5 min
to first workflow run
From log hunting to pointer-first diagnosis
After: 3 pointer hops
Signal-firstreceipt -> pipeline/manifest.json
jobs/<job_id>/manifest.json -> failing_step_events_path
.../events.jsonl -> exact failure evidenceFollow a narrow evidence trail to the exact failing unit instead of widening to broad logs first.
Why use Loom
Pointer-first diagnostics
Open the receipt, follow the manifests, and land on the exact failing events file without scanning broad logs first.
Deterministic runtime artifacts
Receipts, manifests, and event streams give every run the same evidence shape for handoffs, reruns, and tooling.
A workflow engine teams can operate
Validate, compile, run, cache, and diagnose workflows with one model that stays useful after the happy path ends.
How it works
Author
Write a workflow in `.loom/workflow.yml`.
Validate
Run `loom check` to validate schema.
Execute
Run `loom run --local`.
Diagnose
Follow manifest pointers to the failing unit.
Built for agents, grounded in workflows
Structured runtime artifacts let your agent triage failures using the same evidence you would inspect.
Ground your agent in the failing step
Structured receipts and manifests give your agent a stable starting point — no reconstructing context from scattered logs.
Follow stable pointers, not guesses
Pointer paths tie your agent to the same evidence you would inspect, eliminating ambiguity during automated triage.
Your workflow stays the source of truth
Loom is your workflow engine and evidence trail. Your agent benefits because every runtime artifact is structured and deterministic.
Start here
Choose the shortest path for what you need to do next.
Run Hello Loom
Run a first workflow, inspect the artifacts it creates, and diagnose one intentional failure end to end.
See how Loom works
Understand the workflow model, the four-step loop, and how pointer-first diagnostics fit into it.
Diagnose a failing run
Use the diagnostics ladder to go from receipt to the exact events file without widening to broad logs first.
Ready to follow the failure trail?
Run Hello Loom in under five minutes, then inspect the exact artifacts it writes.
Run Hello Loom