Workflows
A Promptev workflow does the same job every time something happens: a form is sent, an e-mail arrives, a file changes, the clock says Friday. You describe it in plain words, test it for real, and publish. It uses your agents where judgement is needed and waits for a person where you ask it to.
How it works
Step 1
Tell the workflow builder the job in plain words: “when the website form is sent, look up the company and score the lead”. It checks what this project can use, names any app you still need to connect, and drafts the workflow: what starts it, each step, who it asks and what one run costs.
No boxes to drag and no code to write.
Then try it before anyone relies on it.
Step 2
Test it runs the workflow once, for real, on the details you give it, and shows what each step did and what it gave back. Change a step by saying what to change, or click one box of the diagram. Publish when it is right; every publish is a version you can roll back to.
You see it work before it runs on its own.
Then it runs every time the event happens.
Step 3
From then on it starts on its schedule, on an event in a connected app, when another system calls its address, or when one of your agents starts it. A step that changes something can wait for a named approver first, and every run shows its path, each step and its cost.
Routine work gets done, and a person still decides what matters.
When it starts
Each workflow has one start, and it starts every time that happens. Nobody has to remember to run it.
What a step can do
In control
Example
The sales team says: “When someone sends the contact form on our website, look up their company, score how good a lead it is from 1 to 10, add a row to our leads sheet, and post the good ones to the sales channel in Slack.” The builder drafts it, names the Slack app the project needs, and shows that one run costs a few credits.
They test it with a real enquiry, read what each step did, and publish. From then on every form starts a run, the research and scoring happen without anyone, and the team only reads the leads worth calling.
Questions
A job that runs on its own, the same way every time, inside one project. It starts on something (a schedule, an event in a connected app, another system calling it, or one of your agents), takes its steps, and asks your agents only where judgement is needed.
No. You describe the job to the workflow builder in plain words and it drafts the workflow, with a diagram you can read. You change it the same way, by saying what to change. The code is there to read if you want to.
A schedule in your time zone; a new e-mail in Gmail or Outlook; a calendar event that changes; a file that changes in Google Drive, OneDrive or SharePoint; a Jira issue or Confluence page that changes; a regular check for new items every few minutes; a call to the workflow’s own web address, signed by Stripe, GitHub or Promptev; or an agent you let start it.
Yes. A step that changes something can wait for an approver you name. They get an e-mail with Approve and Reject, and the request is in Promptev too. If nobody answers in 24 hours, the workflow does what you said it should do when it fails.
It uses credits like an agent: 1 to start, 1 for each agent or model step and 1 for each tool call. The builder shows the cost of one run before you publish, each run shows what it used, and every run has a credit limit (200 by default) so a mistake cannot run up a bill.
It tries again, three times by default, after 1, 5 and 15 minutes. Then it tells the person responsible for the workflow, or pauses the workflow, as you chose. A workflow that suddenly runs far more often than usual pauses itself and says so.
1,000 free credits for a 30-day trial. No card, no seats.