Tutorial
Set up a multi-step workflow in The Loom
Use The Loom when one prompt is not enough. Build draft, critique, research, and review steps, then let the workflow run in the background.
Before you start
Estimated time: 7 min
- An LLMWeave account
- Comfort running single weaves
Open The Loom
1 minThe Loom is where multi-step workflows live. A workflow strings several layers together, each of which can call one or more models.
Lay out your layers
3 minAdd layers for each stage of the task, for example draft, critique, then synthesize. Each layer can fan out across models and pass its result to the next.
Add a human-review pause if you need one
2 minFor anything high-stakes, add a human-review step. The workflow pauses durably, waits for your input, then resumes from exactly where it stopped.
Run it durably
3 to 8 minStart the workflow. It runs in the background with each step checkpointed, so it survives interruptions and you can come back to the result. Cost is tracked per model across the whole run.
Keep going
Rank and fuse multiple model outputs
Generate a few candidate answers, rank them, then fuse the best parts. Use this when quality matters more than speed.
Intermediate · 4 minGround a weave in current sources with web search
Turn on web search when the answer depends on today’s facts: prices, product pages, news, docs, or anything else that can go stale.
Intermediate · 6 minRun a workflow from the API
Call a saved workflow from code, Zapier, Make, or n8n. Start with polling, then switch to webhooks when the run needs to feed another system.