AI Studio
AI that drafts. You decide what goes out.
It writes the template, the reminder and the trade model, and then it stops. Nothing it produces reaches your customer, and nothing it produces moves money, until a person says so.
What it writes
Four capabilities, each one shipped. This list is the product's own feature table, so it cannot name something that does not exist.
Invoice template designer
Describe the layout; the preview is drawn by the real renderer. A draft until a person publishes it.
Pay page designer
Emits colour values, never CSS, and every one is checked for contrast in both themes before it can be published.
Assistive copy
Reminder text, line descriptions and payment-plan explanations. Plain text only — tags and links are refused.
Business model generator
Describe your trade and get a proposed set of terms, presets and invoice fields — every field editable before it saves.
What it is not allowed to do
Six rules, each one enforced by the code rather than by a policy. If we broke one tomorrow the build would go red before the deploy.
A draft is where it stops
Every template, theme, model and line of copy the AI writes is a draft. A person publishes it, or it never leaves the studio.
Schema, never markup
The model answers into a schema and a field outside it is discarded rather than stored. There is no path from a prompt to script, style or markup on your pay page.
What a failed answer costs
Credits are used for every call the model provider bills, including a call whose answer fails validation, so a failed call can still use credits. The gateway retries once and uses a second model only where the platform has configured one. The monthly call count skips an answer that fails validation.
An assistant asks; it never approves
A connected assistant reads and drafts. To charge, send or refund it files a request, and the connection’s mode decides who approves it. In Ask, the default, each request waits for a signed-in person to approve it; Plan is a dry run that saves, sends and charges nothing; in Auto, which only the workspace owner can switch on for a connection (signing in again with a second factor, on a plan with auto-approve rules), the owner’s capped rule approves the requests it covers, and for a connected assistant refunds, voids, card or bank entry and the other never-auto requests still wait for a person.
Your data is not the product
Prompts carry the records the task needs and nothing else. Card numbers are not in the system to send, and a response is masked for digits before it is stored or logged.
What you can switch off
The AI features have one switch for the whole platform, and only a platform administrator can turn it. When it is off the AI callables refuse and the rest of the product is unaffected, because nothing depends on a model answering. A workspace owner or admin can turn off the built-in Agent for their own workspace.
Questions about the AI
Can the AI send an invoice to my customer?
Not by itself. Everything it writes — a template, a pay page theme, a trade model, a line of reminder copy — is a draft until a person publishes it. Sending is a separate action: a connected assistant can only ask for it, and the connection’s mode decides who approves. In Ask, the default, each request waits for a signed-in person to approve it; Plan is a dry run that saves, sends and charges nothing; in Auto, which only the workspace owner can switch on for a connection (signing in again with a second factor, on a plan with auto-approve rules), the owner’s capped rule approves the requests it covers, and for a connected assistant refunds, voids, card or bank entry and the other never-auto requests still wait for a person.
Can an assistant I connect move money?
Not on its own. A connected assistant can read and draft. Anything that charges, sends or refunds becomes a request with a summary the server writes, and the permission that would do it directly cannot be granted to an assistant. In Ask, the default, each request waits for a signed-in person to approve it; Plan is a dry run that saves, sends and charges nothing; in Auto, which only the workspace owner can switch on for a connection (signing in again with a second factor, on a plan with auto-approve rules), the owner’s capped rule approves the requests it covers, and for a connected assistant refunds, voids, card or bank entry and the other never-auto requests still wait for a person.
Can the AI put scripts or styles on my pay page?
No. The model answers into a schema, and a field outside that schema is discarded rather than stored. It emits colour values, not CSS, and every one is checked for contrast in both themes before it can be published.
What happens to an answer that comes back wrong?
Credits are used for every call the model provider bills, including a call whose answer fails validation, so a failed call can still use credits. The gateway retries and tries a second model where one is configured. The monthly call count skips an answer that fails validation.
Can I turn the AI off?
Not per workspace. The AI features have one switch for the whole platform, and only a platform administrator can turn it. With it off the AI callables refuse and the rest of the product is unaffected, because nothing in the billing path depends on a model answering. An owner or admin can turn off the built-in Agent for their own workspace.
Let it draft the first one
Describe your trade and it proposes the words, the fields and the terms. You edit any of it before it saves.