Steps

A wizard runs its steps from top to bottom. The last step is always the result.

Pages

A page has a title, its fields (a good page has at most five) and a button. Each field has a kind:

A text field can have a “Scan” button that fills it from a QR code or barcode.

A field can start with a value an earlier step proposed. A line-items field, for example, shows the positions an AI step looked up, for the person to correct.

A field can be shown only under a condition (“Show only when”): “Street and town” only when the person chose delivery, “Which allergy?” only when they ticked “Allergies”. The condition can read another field of the same page, and the field then appears as the person answers. It can also read anything known before the page. A hidden field is not required and is not kept.

The choices of a choice field can come from earlier data (“Choices from earlier data”) instead of a fixed list: the suppliers a research step found, one column of a table it made, or a column of a list the wizard keeps. The fixed choices stand in while that data is empty. When the choices are the rows of a table, each can show as a card with a picture, a line or two of text and “Details” to open its page.

An appointment field takes its times the same way, usually from an AI step that looked up the free times in a calendar. The times show in the calendar’s own time zone, grouped by day.

A field can also be left to the AI. The AI reads what was answered and found so far, and shows the field when it finds a statement true (“The complaint is about a damaged article”). Fields that are alternatives of each other form a group, and the AI shows the one that fits (“How should we reach the person?”). The AI decides once, when the run reaches the page. A page whose fields the AI chooses holds up to 12 of them and shows at most five.

AI steps

An AI step has instructions and the tools you allow it:

Every AI step can also read and write the wizard’s lists, list its files and read documents: PDFs, scans and photos, Word, Excel, CSV and saved mails. Where the space has Knowledge, the step searches it too. Every AI step also knows the space’s Info & brand: its title, what it says about itself, its facts and colours.

Further tools come from three places:

• The person’s accounts: connectors the wizard uses, where the person connects their own account during the run. The step gets the service’s actions as tools. Reading is free; before anything changes in the person’s account, they are asked. See Connectors.
• Connected systems (MCP servers) of the space, where you allow them on the step.
• Plugins installed in engenty wizards. See Plugins.

Model class

A step names the kind of model it needs, never a model:

On a Mac, Classifier and Low can run on Apple Intelligence. It is free and stays on your computer, but it handles only short texts and no tools. There, give a step that calls an API or reads documents the class Standard.

Images, video, voice notes and voice-overs have classes of their own. You choose which model serves a class under Models and account. “Thinking effort” sets how much effort the model puts in. Leave it on “Automatic” unless a step is clearly trivial or clearly hard.

What a step hands on

A step’s output is text, Markdown or structured data with named fields: a text, a number, a list, a table, a yes/no answer, a choice or a score on a scale (“low, medium, high”). Later steps use these, and branches decide on them.

Generate

An image or video step can make one result per entry, at most eight: per uploaded photo, per image of an earlier step, or per row of a table. In a review the person can have single results made again.

Views

A view shows what the run found in the wizard’s own look, on any screen: key figures, a list of facts, a list of results with pictures, a table or a simple chart (bars, a line, a ring). It is built from a fixed set of parts that the assistant writes once, and every run brings new data. It costs nothing to show.

A view can also leave some of its parts to the AI, for example a chart only when there is something to compare. And an AI step can lay out a view of its own result, which reviews and the result page then show instead of a plain table.

For anything the parts cannot show, such as a map or an animation, use a widget.

Widgets

A widget is an interactive HTML view written once into the wizard’s files. Every run only brings new data, so it costs nothing to show and looks the same every time. A widget can also be a film: its timeline is then rendered to an MP4 with sound. Ask the assistant for one in the chat.

Films

A film step hands a brief, the run’s material (voice-overs, pictures, uploads, texts) and the space’s brand to the AI client installed on your computer (Claude Code). The client writes the film as code, looks at its frames and fixes them. engenty wizards then renders the MP4: 16:9, 9:16 or square. A film step runs only where Claude Code is set up with your subscription, never on credits or in the cloud, and it takes 5 to 40 minutes. The voice comes from a generate step before it.

Reviews

A review shows what earlier steps made. Depending on its settings, the person can edit the text (“Text can be edited directly”) or ask for a new version with a note on what should change (“Ask for a new variant”). A list shown in a review is edited in place. A list with files is checked row by row, each row beside its document.

The result

Each entry of the result page names a step and the formats the person can download:

Under “Send & publish” you choose where people can send or publish the result with one button, besides the link and the downloads. Each target shows its state: “Ready”, “AI agent · only you” or “Not set up”. Switch targets off, or prefill their forms with “Insert”, and the result keeps its own list; “Reset” brings the default back. See Share a result.

When you choose the result (the flag), the preview shows its result page, filled from the last test run. “Visitor” and “You” switch between what each of them gets. A target opens to its form as the visitor sees it, with the test run’s answers in it; nothing is sent from there. The result node on the flow shows the targets too: filled for visitors, a ring for you only.

Branches

A step can send the run somewhere else when a condition holds: a field is or is not a value, is one of several, is filled in or empty, is greater or less than a number, or contains a value. A condition can also read what an AI step found, how sure the AI was of a decision, or how many rows a list the wizard keeps has. Several conditions on one branch must all hold. The first branch that matches wins; otherwise the step’s “otherwise” branch, or else the next step, follows. The diagram shows the branches with their conditions.

A branch can also be decided by the AI: instead of a condition it has a statement (“The person wants a refund, not an exchange”). When no condition matched, the AI reads what was answered and found so far and takes the branch whose statement is true; if none clearly is, the step’s “otherwise” branch, or else the next step, follows. All such branches of a step cost one small decision together.

A branch back to an earlier step makes a loop, for example “write, check, write again until the check passes”. A loop stops after the number of times you allow, at most ten if you set nothing. The step that runs again sees what it made the last time.

An AI step can also work through a list: once per row of a table an earlier step made, or per row of a list the wizard keeps, at most 20 rows. Its results come together in one table.

What a wizard keeps

• Lists hold rows between runs, separately for each person who runs the wizard: the receipts already booked, the customers already written to.
• Shared lists hold one set of rows for every run: requests, sign-ups, a log. The person running the wizard never sees them; you find them under Space → Data, below the wizard, and in the editor under “Memory”.
• Connected accounts and sign-ins are kept for the person’s next run.

The person sees and deletes all of it under “What this wizard remembers”. See Run a wizard.

AI media

Images, clips, voice-overs and films a model made or changed carry a marking in their file, and every page lays an “AI-generated” or “AI-modified” label over them (EU AI Act, Art. 50). Nothing is drawn into an image. A film shows the label in the picture, since it leaves the page as a video file.