Route each alert to one clear next step
An alert can mean very different things. Customers may feel it now. The severity sets how urgent it is. A runbook may already cover the problem.
The first question is whether customers feel the impact. If they do, the alert goes to a person at once. If not, the flow asks about severity.
Lower-severity alerts ask one more question: does a runbook exist? A known runbook sends the agent to follow it and report whether the alert cleared. If it did not clear, a person is paged. An unknown alert becomes a ticket.
Each outcome names its next step. The agent does not invent a plan, and no path leaves an alert without an answer.
The flow in one picture
Alpha, private
Try this with early access
Rein is a proof of concept. Syntax and tools can still change. Request early access by email, and tell us which decision flow you want to build.
Related use cases
- Decide the next step for each pull requestDecide whether a pull request merges, gets a review, or waits for CI.
- Answer tickets, or hand them to a personAnswer a support ticket, start a refund, or hand it to a person.
- Pick the right model for each delegated taskDelegate each task to a subagent, with the model that fits it.
- Ask a person, wait, then branch on the answerAsk a person a question, wait for the answer, then branch on it.