Not a black box. Here's exactly what runs, in order.
Every Rivet automation is built from the same four stages. No custom software, no long build cycle — we configure a pipeline against the tools you already use, test it against your real data, and hand it over running.
The pipeline
Using maintenance request triage as the example — the same shape covers reporting, follow-ups, and quoting.
This runs on n8n (a workflow platform) — not custom-built software. That matters for you: it's inspectable, it's not locked to us, and it can be handed to any developer later if needed.
What happens at each stage
Trigger
A workflow watches a specific inbox or label — never a whole personal inbox — and fires the moment a matching email arrives. No polling, no delay.
AI classification
The email text is read and turned into structured data: category, urgency, unit or reference number, and a plain-language summary — plus a drafted reply.
Logging
Every request is written to a spreadsheet automatically — the record an owner report gets built from, with nothing manually typed in.
Action
Depending on what you choose: the reply sends automatically, or it's held in a draft folder for a human to approve first. Most clients start with the second and loosen it once they trust it.
Where a person stays in control
Nothing sends without a rule you set
Auto-send is opt-in per category. An emergency plumbing issue can be routed straight to a human, every time, while a routine request auto-replies.
You can see every decision it made
The logged spreadsheet is the audit trail — what came in, how it was categorized, what was sent, and when.
It's built on tools you can walk away from
n8n, Gmail, and a spreadsheet — all things you already have access to or can get in minutes. Nothing proprietary holding your data hostage.
See stage 2 running for real
60 seconds — a maintenance request comes in, gets classified, and a reply gets drafted. No one touches it.