The pipeline in your head. Claude
runs it end to end.
A recipe is a typed chain of actions (source leads, verify, enrich, draft, send) with cost estimated up front and every step logged. Build it in the Coil dashboard, or describe it to the AI you already use. Coil runs the graph and reports back. Below is the agent view.
Source
Verify
Enrich
Draft
Send
Three recipes your AI can save today.
You don't build them. You describe them once, the agent saves the shape, and you re-run with new inputs whenever you want.
Weekly net-new from a saved ICP
Funding-event triggered outreach
CSV → personalized sequence
Cost before run. Cost after run. No mystery.
Every step has a typed provider and a known price. The agent estimates the bill before you approve and records the real numbers after. You see the line-item cost of a verified personalized 500-lead campaign before you commit, and the receipts match the estimate.
Recipe questions, asked early.
The ones you'd ask in a 15-minute call. Answered here so the call can go somewhere more useful.
Do I have to "build" a recipe?
No. Describe the whole campaign in chat. The agent picks the steps, the providers, and the order. If you want to override (say, "use Clearbit instead of Apollo for enrichment"), the agent updates the recipe before running.
What if a step fails halfway through?
Partial progress is preserved. The agent tells you which leads completed each step and offers to retry just the failed ones. Recipes are restartable, not all-or-nothing.
Can I edit a saved recipe later?
Yes, in chat. "Edit my Q2 recipe to skip enrichment." Or "add a step that writes to my HubSpot before send." The agent shows the diff before saving.
Where do provider keys live?
Encrypted in your org settings. You set them once; the agent uses them when a step needs them. Coil never sees the plaintext after first save.
Describe the whole graph in one sentence.
Type it in the dashboard or in chat. Coil runs the steps. You get the receipts either way.