Skip to main content

An AI website builder for agencies

An agency's problem with a website builder is rarely the first site. It is the twentieth: twenty logins, twenty billing relationships, twenty slightly different stacks, and a first draft that costs the same on the twentieth site as it did on the first.

Aglyn AI runs from the console your team already signs in to, against the sites your organization already owns. One brief can start a build on many client sites at once — and each one stays a separate job you review.

Rolling out

Aglyn AI is a release-flagged feature, currently being rolled out — it is not available in every workspace yet. This page says how a batch behaves, and grows with the feature.

One brief, many client sites​

A batch takes the brief once and starts a site build job per site you picked. It is not one job writing twenty sites: each site gets its own plan, its own confirmation and its own drafts, so a plan you dislike on one client costs you that client's credits and nothing else.

The batch also asks once where form submissions should go, and binds that answer to every job it starts — the Inbox, or the Inbox and a CRM lead for each message that carries an email address. You can change it on any individual form afterwards.

What is shared, and what is not​

The brief is shared. Nothing else is.

Each job reads the site it is building for — that site's theme, its layout, its components and its forms — and builds from those, following the building rules. Two clients with the same brief and different themes get drafts that look like their own sites, because the generator has each site's own material to build from rather than a prompt and a blank page.

The organization model it assumes​

A batch operates on the sites inside one organization. If your agency runs every client as a site in a single workspace, a batch can reach all of them; if each client has its own separate organization, a batch reaches one client at a time.

Which of those to run is a decision that predates AI and outlives it — running an agency workspace covers the trade-offs, and AI does not change them.

Keeping spend separate per client​

Credits are a workspace pool, so without a limit one busy client can draw down what another's work needs. An allotment is how you divide it: a monthly number of credits for a whole site, a member, or a site collaborator, counted from the first of the month.

A hard allotment stops AI requests for that site once it is used; a soft one records the same number and lets work continue. Either way an allotment is a share of the pool and never an addition to it — the workspace's own limits still apply first. AI allotments, usage and model choice has the full behavior, including limiting which models a client's site may use.

Turning AI off for one client​

A client who does not want generative work on their site can be opted out without affecting anyone else. Switching AI off for a site hides the assistant, Describe it, the AI cards and the editor's AI controls on that site, refuses every AI request made for it, and stops its queued jobs without spending credits.

It does not stop the workspace's AI add-on, its credits, its allotments or its billing, and AI keeps working on the workspace's other sites.

White-label is a separate setting​

Generating a client's site and removing Aglyn's name from it are two different features. White-label replaces the Aglyn brand across the console, your published sites and transactional email; it is its own setting with its own entitlement, and having the AI add-on does not include it.

Who publishes​

Nobody's client site goes live because a job finished. Every generation writes an unpublished draft, or a new unpublished version of something that already exists — it never flips a published version, registers an address on a live site, changes navigation or sends anything.

An account manager opens each draft in the editor, changes what they want, and publishes it. On twenty sites that is twenty deliberate acts by a person, which is the point: the batch removes the typing, not the review.