Skip to content
Product · automated workflows

The request you repeat, run as an app.

Same inputs, same steps, same shape of answer, and a review step where a person still signs off.

One request fans out into parallel runs over the same approved sources, and each one comes back with its own citations rather than one merged answer nobody can check.
Your approved sources only A person releases the output Every line cited
Three running at once

Different work, the same shape

Three packaged apps from three different practices. Note what they have in common: named inputs, a fixed sequence, a stop for a person, and a cited artifact.

Why an app and not a prompt

Four things a packaged app gives you that a prompt does not

The tenth time should not cost what the first did

The request you get every week has the same inputs, the same steps and the same shape of answer every time. What changes is the file. That is a workflow, and it is currently being done by hand by whoever is free.

A person still signs it

Every workflow has a review step, and it is not decoration. The app stops, shows you what it found and what it wants to do, and waits. Anything that would go out under your name goes out because you released it.

The deliverable arrives cited

Not a summary you then have to verify line by line. Each statement in the output names the document and page behind it, so checking it is reading rather than re-doing it.

It is an app, not a prompt

Your team opens a named app with the fields it needs and presses run. Nobody has to be good at asking, nobody has to remember last month’s wording, and the tenth run is the same as the first.

Questions we get first

Before you hand over a workflow

Is this a chatbot with extra steps?

No, and the difference is the interface. A packaged app has named inputs, a fixed sequence, a review step and a defined output. There is no prompt to get right and no wording to remember, which is exactly why the tenth run matches the first. Chat exists too, and it is a different tool for a different job.

What happens at the review step?

The run stops and shows you what it has found and what it proposes, the substitution, the flagged inconsistency, the item with no match. You approve, adjust or stop. It does not proceed past that point on its own, and that is deliberate rather than a limitation we are working on removing.

Who builds the workflow?

We build the first ones with you, from the requests you already repeat, and your team can build more. The Marketplace also carries installable apps and templates so you are not starting from a blank canvas.

Can it run over documents I have not approved?

No. A workflow reads the knowledgeBases you put in scope and nothing else. The scope is a decision you make and can inspect, on every run.

Start with one you already repeat

Which request does your team do every week?

Bring that one. Twenty minutes, and you will see it run on your own documents with a stop where your review belongs.