A hosted project's page — live preview, the Assistant window, Overview, Feedback, Requests, Changelog, and Hosting

At the top, the same facts the list row showed — thumbnail, status pill, client/kind/category, live address and hosting badge, last-activity line — repeat here, so clicking in from the list never loses information you already had. Once the project is live, a "Copy live link" button sits right next to the address (2026-09-15) — the fastest way to grab the exact URL without opening "Send this to your client" further down the page. A "← All hosted projects" link returns to the list.

Right below that is the project deck. First the project's goal in one line (from the brief). Then six doors, one row: Site plan (its own page — see "The Site plan" guide — the pages, modules, business facts, brand, decisions, and open questions this project actually has, kept current automatically), Brief & rules (the Assistant's Brief & rules tab: what the AI knows and follows), Kit (the project's own Kit tab: the modules and business tools this project has, plus a link into the deeper Kit Explorer to browse and add more — see "The Kit: browsing, previews, and adding a module"), Versions (every change, undoable), Requests (submissions from the site's own forms, with the count of new ones), and Feedback (pins and comments, with what is waiting on aiKit or on the client). Under the doors, a collapsed Technical details drawer summarizes where the site is hosted and how long the free preview has left; open it for the full grid: Hosted at (aiKit hosting on a temporary aiKit address, aiKit hosting on a custom domain, or hosted elsewhere), Address, Domain check (with when it was last checked), Started, AI edits (how many AI edits the client has left and how many more are allowed today: the free preview gives 10 edits, at most 3 per rolling 24 hours, and an aiKit admin sees an Add 5 edits button that lifts the daily limit because paid edits are not rationed), Free preview (the 90-day window from the project's start, with the end date and days left; hosting stays free after it, AI edits are metered), Last build or deploy, Last activity, Kind, How it started (with the source address where one applies), Build tier (Standard or Premium), Client edit link (whether the access code that powers Edit on the live site and the feedback overlay exists yet), and Feedback mode. Last, Linked projects: the parent project, the projects that are part of this one, other projects for the same client, and a link to the list filtered to that client.

• Overview — the live-site review link, Quick edit, cloning, the client link, the Project details editor, Add to this project (business tools and side projects), the Project brief, the Build plan & files card, and Team access.
• Kit — what this project has from the kit and the catalog to add more, at full width (the deck's Kit door opens it). The left column lists every module on the plan with its real state from the last build: Built (with an "Open on the live site" link once the project is live), Not in the build (the last build did not produce that page; ask the Assistant to add it or rebuild), or Planned, each with the customization note written for it, and the business tools alongside as work items. The right column is the same quick-add catalog the Assistant's Kit tab shows, with previews, notes, and "Add to this project" (which records the pick on the plan and files a work item for our team). An "Explore the full kit" banner above that quick-add list opens the deeper Kit Explorer, its own full page — see "The Kit: browsing, previews, and adding a module" — with every category, a detailed preview of each module (a real rendered page for a built one, an honest illustration for a planned one), and a fuller add panel with suggested customizations, a notes field with attachments, and "Add to plan" vs "Add and build now."
• Feedback — every request from the client or from staff, with its full status lifecycle.
• Requests — quote requests, contact messages, and booking requests that visitors sent through the site's own Get a quote, Contact, and Book online pages (aiKit's tested modules). Each one is a real lead with a status, attached photos, and a private team note; the tab's badge counts the ones still marked New.
• Chat — one running, page-wide conversation with the AI about the whole project, not tied to a specific spot on a page.
• Changelog — every change that's actually shipped, in order, with a client-shareable proof link.
• Hosting — the full plain-language tutorial on where the site lives and how to connect a real domain (its own guide-length tab, summarized below).

For a project still on its very first build, a banner sits at the top of Overview showing which stage it's in — Build queued, Importing their current site, or Building the first version — with a spinner, and a note that this runs in the background so it's fine to leave the page; nothing goes live until someone clicks Go live regardless of how far the build has gotten.

If the first build fails instead, the banner switches to "The first build didn't finish," explains that nothing is lost (the brief stays saved below), and offers a "Retry build" button. Either way, a collapsible "Progress log" underneath shows the real build output — open by default when it's failed, so the reason is visible immediately.

Once a project is live, the workbench's header is "Review & annotate the real site": "Open live site to review →" opens the actual live site with the feedback overlay turned on, where anyone can drop a pin on any page, dictate feedback, attach a screenshot, or ask the AI to fix a pin directly — and "Redeploy" pushes the project's current files to production again, for a change made by hand in code rather than through the AI editor.

Before a project is live, the same spot instead shows a "This project isn't live yet" (or "This clone isn't live yet") card explaining that Quick edit and Chat both already work on the real files even before launch, and that Go live is a separate, deliberate step you take once you're happy with it — with the Go live button right there.

Quick edit, right below, is a plain-language box: type a change ("make the phone number bigger"), optionally dictate it or attach a screenshot, and "Request change" logs it as a real feedback item and immediately asks the AI to make that exact edit to the real code — redeploying it live right away if the project is already live, or just saving it if not. A status bar tracks progress (working → done or failed) with a "view live" link and a hard-refresh reminder once it's done, and a "This isn't actually fixed" button reopens it if the fix didn't actually land.

"Clone this project" copies this project's real files into a brand-new project — for a new client who wants to start from the same foundation. You name the new project and can add any number of distinct "what should be different" change requests one at a time (each its own item, removable individually), plus an optional reference screenshot. Every change you list is actually applied to the new copy's real code right away as part of cloning — it isn't just noted for later — though the clone still isn't public until someone separately clicks Go live on its own page.

"Send this to your client" (2026-09-15) gives you four separate, clearly labeled links instead of one — none of them need a login on the client's end. "Copy client feedback link" (shown first, once live — it's the one most people actually want) copies the live site's own address with its feedback/edit overlay already turned on (the same ?aikit-edit=code parameter the "Edit on the live site" button uses — see "The live preview above the tabs" below), minting an access code first if the project doesn't have one yet, so a client who opens it can pin feedback right away. "Copy live site link" copies the plain address with no overlay — for when you just want them to see the real, working site with nothing else turned on. "Copy questions link" copies the same link the Questions door sends, so you can hand it out again without opening Questions — it mints the same kind of access code if one doesn't exist yet. "Copy updates link" (the original button, renamed for clarity — it used to be the only one, called just "Copy client link") copies a lightweight, login-free link to this project's recent updates, for pointing a client at what's shipped without walking them through the full Client Hub. The live address in the identity header above carries its own "Copy live link" and "Copy client feedback link" buttons too, for when you don't want to scroll down here.

This is what the Browse-by-client sidebar on the list is actually built from: Client (a dropdown of every client, plus an inline "Or add a new client" name field with its own Add button so you never have to leave this page to create one), Kind of project, and whether this project is "Part of a bigger project" — that picker only offers other projects belonging to the same client. A Category and Subcategory pair (free text) is the alternative way to browse the same list. A Display name field overrides what shows on the card when it should differ from the client's own name, and a Client email field makes the project findable by that address later.

For a project that came from an import or transfer, a "Their current site" URL field also appears here — the same source address shown under Technical details.

"When a client submits a request" is a two-way toggle: "⚡ Straight to work" (the default — a submitted request is applied and redeployed immediately) or "🔒 I review first" (a submitted request waits in the Feedback tab until someone clicks "Approve to work" — that approval is what actually starts the AI on it). "Save details" writes everything on this card; a brief "Saved ✓" confirms it went through.

Any project can grow after it exists. This section has two cards. "Business tools kit" opens the same kit browser the wizard uses, limited to the business-tools categories (marketing automation, invoicing and billing, customer records, quotes and proposals, staff tools, reporting, integrations, loyalty, support). "Add to the site" opens the site-kit categories instead (pages, forms, galleries, store, booking, SEO). Picking a module does two things at once: it is added to this project's plan (it appears in the list on this card and under the Project brief card below), and a work item is filed in the Feedback tab as "Our team", so it shows as Waiting on us, appears on the Feedback page across projects, and can be run with "Fix with AI" or discussed like any other request. Picking the same module twice is refused with a note that it's already on the project.

"Side projects" is for a separate project for the same client that belongs under this one — an app, an admin system, marketing automation, another site. The buttons are one per kind (the current project's own kind is left out). Each opens the New project wizard with the client, the parent project, and the kind already filled in, so the wizard starts at the project name. Existing side projects are listed above the buttons with links to their own pages; they also show as sub-projects in the Browse-by-client sidebar.

For any project created through the wizard (or otherwise carrying a saved brief), this card shows every answer given during Create-a-project: the goal, audience, what to keep, brand assets and attachments, current domain, deadline, and any other notes — each rendered under its real question label, with attachments shown inline (an image thumbnail, or a download link for anything else). Beneath the answers, the Site kit and Business tools kit picks each list their module name and the client's own note on what should be different about it, and a final line states the hosting choice made in the wizard ("On aiKit (default)" or "Client hosts it themselves").

"Copy brief" copies the whole thing as plain text — the exact same format the wizard's own "Copy plan" produces — ready to paste anywhere.

The plan is a living document, not a changelog. After every change that lands on the site (a chat edit, a pin, a circle), aiKit rewrites the parts of PLAN.md the change touched, so the page list, the modules, the business facts, and the decisions always describe the site as it is now. It never grows a list of dated amendments; the Versions tab is the history. One line under the title says when it was last brought up to date.

For any project that came out of Create-a-project (or was brought in from another tool), this card sits under the Project brief and shows three things the build engine can prove rather than claim. First, the Site kit checklist: every module picked in the wizard is looked for in the real files after the build finishes (each one has to carry a small marker the build was told to add), with a pill reading "N of M built" and, per module, either the file it lives in (click to open it) or "Not found in the output — needs a person." Anything missing was already re-requested from the AI once automatically before it got flagged here. Business tools kit picks are listed separately as planned work items — they're never faked inside a static site.

Second, the build's own PLAN.md, rendered as readable text: the pages built and why, a Modules built section, Planned business tools, the brand decisions, and open questions for the client. A build must write one — if the AI skips it, a short follow-up pass asks for it, and if that fails too the engine writes a minimal, honest one itself and says so in a pill next to the heading, so a plan you're reading was never silently invented.

Third, the real file list from the engine's copy of the project, each with its size; clicking any text file opens it read-only in a viewer (binary files and the build's own context folders are never shown). Projects linked from existing code or cloned don't have a plan, and the card says so plainly instead of showing an empty box.

When a project's Site kit includes the "Quote or Estimate Request Form," the build doesn't ask the AI to invent a form: aiKit's own tested module renders the Get a quote page (its fields, urgency choices, and whether it asks for an address or photos are filled from the brief; contact details are only ever real ones from the brief or the mirrored site, never invented), and the AI is only told to link to it from the navigation. After the build, the engine re-generates that page with the site's real header, footer, and stylesheet so the form itself can't have been broken.

On the live site, pressing Send stores the request on aiKit immediately — it works as a plain form with no JavaScript, on any host, with no email setup. The Requests tab lists each one newest first: name, phone (tap to call) and email, the service and how soon, the address if asked, the message, and which page it came from. A status dropdown (New, Contacted, Won, Closed) and a private note box per request keep the team in step; the tab badge and the list row's "N new requests" link both count what's still New. Photos are stored with the request (the form shrinks phone photos before sending, up to four per request) and show as thumbnails on the row — click one to open it full size. Two more tested modules land on the same tab: the Contact Form That Routes Right (tagged "Message" with the topic the visitor chose) and the Book online (tagged "Booking" with the requested day and time, shown in the visitor's own time zone). A booking holds its slot on the live site until it's marked Closed, and the booking page refuses a slot someone else just took. A new request of any kind flips the project to "Waiting on us" on the list. The client can see the same leads themselves: "Copy client link" at the top of the tab gives a page (aikit.ai/p/<project>/requests, also linked from their Recent updates page) that asks for the project's access code — the same code they use to leave feedback — and then shows every request with contact details, photos, and a status they can change, but never your private team notes. A project with no access code set leaves that page open, so set one before sharing the link.

A separate list of real aiKit accounts with access to this specific project — distinct from the shared access code clients use on the live feedback overlay. The owner can't be removed; anyone else can. Adding someone takes their email — they need their own free aiKit account first (created the same way as any /start/classic account) before they can be added here.

Every item — a pin dropped on the live site, or a Quick edit request — carries the same lifecycle: Open → In progress → Fixed, awaiting client review → Confirmed, with a separate Reopened state if a "fixed" item turns out not to be. "Waiting on us" means Open or In progress; "Waiting on client" means it's marked fixed and they haven't confirmed yet. A "?" button expands a full explanation of every action inline, right on the tab.

Each item shows the message, what it's about (the element it was pinned to, if any), an attached screenshot if there is one, a status pill, and a role tag — "Client" or "Our team" — making clear who actually filed each request. Available actions depend on its current status: in review-first mode, an open item shows "Approve to work" first; from Open/Reopened/Approved/In progress, "Fix with AI" sends it straight to the AI editor, which edits the real code, redeploys, and marks it fixed with the live link as proof — the normal path. "Start work by hand" marks it in progress without the AI, for a change made directly in code (pair with Redeploy on Overview once it's live, then Mark fixed). "Mark fixed" asks for a note or URL proving it's done. A fixed item can be "Confirmed on client's behalf," and anything can be "Reopened" if it isn't actually right.

Under each item, a "Work done" list shows every shipped change tied to it (with a link to view the live page and a link into the Changelog) plus its verification note or URL if one was given. A "Discuss" toggle opens a threaded reply conversation on that one item — the client sees the exact same thread from their own link.

Opening this tab from a link on the cross-project Feedback page (or anywhere else with a #fb-<id> link) scrolls straight to that item and briefly highlights it, so following a link from elsewhere never leaves you hunting through a long list.

Above the preview, "Edit on the live site" opens the real site in a new tab with the editor turned on (the project's access code rides in the link once and is remembered for that tab): an "Edit with aiKit" button in the corner opens the Assistant beside the page with Chat, Pins, and What changed (the list of changes; the restore buttons are staff-only on the dashboard), pins land on the exact element you click, Draw a line marks any area, and a landed change reloads the page. The bar above the dashboard preview also carries shortcuts: Site plan, Versions, Kit, and Brief & rules, so the things you reach for most are one click from the site itself. Public visitors never see it. The first time a built project is opened, a note above the preview says the first build is a starting point, not the finished site, with a Start improving it button that opens the Assistant; Got it hides it. Right under the project's identity header, the live site itself is shown in a frame (once the project is live; before that the block says so and the Assistant still works on the code). Above it: Desktop / Tablet / Phone width toggles (the same idea as the old creative tool's preview toolbar; the frame narrows to 768px or 390px), "Pin a spot" for AI feedback on an exact place, a reload button, "Open" in a new tab with a fresh cache-busting address, and "Ask the assistant", which opens the Assistant window on Chat. When the Assistant is expanded to a split screen, this same preview moves beside the chat.

The Chat tab is gone; in its place a floating "Assistant" button sits at the bottom-right of every project page (a row's "Chat" quick tool on the list opens it directly). It opens a panel docked to the right (drag its left edge to resize it) or, with the expand button in its header, a split screen with the live site on the left and the panel on the right. Esc or the × closes it and focus returns to where you were.

Chat is a plain conversation about the whole site. Type or dictate, attach up to four images, press Enter. Signed-in staff also get commands: the ∞ button beside the microphone opens "Your rules and commands", where Standing rules (a few lines that load into every project you chat with, on every request; the Brief & rules wins if they disagree), the ∞ command (what runs when a message ends with ∞; a built-in version checks the whole site for consistency and reports what changed until you write your own), and Slash commands (a short name and what it does, up to 40) are kept. Type / in the box to see your commands; pick one, or press Enter on the top match, and it expands into the box so you can add to it before sending. "+ ∞" adds the symbol to the end of what you typed. The client's own chat on the live site has no commands. A question gets an answer; a request for a change is made in the real code by the same edit loop the rest of aiKit uses, and while it works the panel shows a working card instead of a spinner: a clock, a plain-language stage ("Small changes usually land in under a minute"), the steps already done ("Searched the site for (555)", "Edited contact/index.html"), and the current step. For a change that appears on several pages (a phone number, hours, an address) the assistant searches the site and edits each spot in place rather than rewriting whole pages, so most changes land in under a minute. Sending stops the microphone if it was on. Only one request runs at a time per project; a second send while one is running is told to wait. If the project is live, the change deploys and the preview above reloads; if not, it is saved to the code until Go live. Every change becomes a version (see Versions). The assistant reads the project's Brief & rules before every reply and follows what it says. The chip in the panel's header counts AI edits: every client gets 10 free edits shared across their projects (a change the assistant actually makes; a question costs nothing), at most 3 a day, and when the day's or the client's free edits are used the chat says so and offers "Ask for more credits", which files a request for the aiKit team. Paid edits are not rationed by day.

Pins is AI feedback on an exact spot, and Draw is the same for an area: press "Draw" above the preview (or "Draw a line" on the Pins tab, or in the panel on the live site) and drag a straight line across anything with the mouse or a finger; the line pauses so you can Undo and redraw it if it's not quite right, or "Use this area →" to confirm it. The line's (padded) bounding box becomes the pin's target area, the AI is told which element sits under it and how big the area is, and on the live site an orange line stays where you drew. "Ask AI to do it" really runs the change (2026-09-11): the pin turns to Working on it, the AI edits the real code and redeploys, and the pin turns to Fixed with a View fix link to the live page; if it could not be done, the pin goes back to open with the reason in its history (the engine itself writes that, so a long edit can never leave a pin on Working on it; after 15 minutes without a result the list shows Taking longer than expected with an Ask AI again button). Click "Pin a spot" above the live preview, click the place on the site, describe the change, and choose "Ask AI to do it" (the change is made and the request is marked Fixed, awaiting review, with a link to the live result) or "Save for later" (a normal open request). Open requests from the client show here too, each with its own "Ask AI to do it". Feedback in words (2026-09-11): paste what the client said (an email, call notes, a transcript), choose From the client or My own notes, and press Turn into requests: the AI splits the text into one open request per change, each aimed at the page it belongs to, and the original text is kept under Notes in Brief & rules. Nothing on the site changes at that point; each request still runs through Ask AI, turns to Working on it, and ends as Fixed with a View fix link, so you see every change start and finish. Save as a note skips the split and keeps the text as background only. Splitting costs tokens (it shows as "split" on the Usage page) but never an edit.

Redesign this site (2026-09-12, staff only) is a whole-site visual redesign, not a feature request: press "Draft a direction" on the Pins tab (or the "Redesign this site" door on the project page) and the AI picks a small set of your approved reference sites — real pages you starred as design references, matched to this project's industry — reads their look and feel, and writes one concrete art-direction brief: a named color palette with real hex values, a type pairing, the section rhythm, several named signature sections, and rules for motion and icons/illustration. "Choose references" opens the same reference-gallery picker the planner uses: search or filter by industry, pick up to 6 sites yourself instead of the automatic match, or leave none picked to keep the auto-match. Read the brief, edit any line, or press "Try different references" to redraft. "Run redesign" sends that brief through the same edit loop as any other change — every page, not just the home page — and costs more than a single edit and usually takes 7-11 minutes, both stated up front on the button. While it runs the card shows a live elapsed clock with an honest, plain-language description of what's happening at that point, not just a spinner; a past run is listed underneath with the brief it used, the reference sites it was grounded in, and a before/after screenshot of the homepage once it lands, so you can see the actual difference rather than take the AI's word for it.

Versions lists every version of the site, newest first: the Original (recorded before the first change), the first build, each chat edit, each pin fix, and each restore, with who made it, when, and which files changed. "Restore" on any older version asks once, saves the current state as "Before restoring", puts the older files back, redeploys if the project is live, and records the restore as a new version, so a restore is itself undoable.

Brief & rules is the living document the AI reads on every build and edit. Sections: Business facts, Brand, Rules & instructions, Content guidelines, Approved styles, Decisions, Notes, Learned. Notes holds plain notes from calls, meetings, and emails, plus feedback pasted on the Pins tab; the AI reads them as background and nothing under Notes changes the site by itself. Add an entry in plain words: choose the section, choose whether it applies to every project for this client or only this one, type or dictate, Add. When the assistant learns something durable in conversation it files it under Learned as unconfirmed; the panel asks you to Confirm or Remove each one, and only confirmed entries count as the owner's own statement.

Kit is the catalog of self-contained modules. "Built" shows the tested modules the engine renders as real pages (Get a Quote / Request Service is the headline; Contact form; Book online), each with a plain summary, features, why it matters, how it customizes, a Preview that renders the real template with sample data, and a customization-notes box. "Whole catalog" adds the planned modules that become work items. "Add to this project" records the pick on the project's plan and files a work item, the same as the Overview's kit browser. The same catalog also has its own full-page Kit Explorer (at /projects/&lt;id&gt;/kit, or browse-only at /projects/kit with no project yet — see "The Kit: browsing, previews, and adding a module") for a deeper, illustrated browse: categories, a full detailed preview of every module, and a fuller add panel with suggested customizations, your own notes and attachments, and "Add to plan" vs "Add and build now."

Questions (2026-09-14) is for the discovery questions a new OR existing client should answer, separate from Pins (feedback on something already built). Unlike Chat, Pins, Versions, Brief & rules, and Kit, it isn't a tab inside the Assistant panel — it's its own full page, reached from the project deck's Questions door or a row's More ▾ menu on the projects list, since drafting and reviewing a real batch of questions needs more room than a side panel gives it.

The brief is a running, saved thread, not a single box you type into and lose: add one note at a time (type or Dictate, then "+ Add to the brief"), each lands as its own entry with a timestamp and a remove button, and it's all there again next time you open the page — nothing lives only in the browser tab. Come back tomorrow, or in a week, and add more; it all counts.

When there's enough, press "Draft questions from everything above": the AI reads the whole thread, plus anything already confirmed in Brief & rules so it never re-asks what's on file, and proposes a set of specific questions — usually around 15, more than a short marketing-site questionnaire would need, because the goal is real detail, not a token gesture. Each question is either a fill-in question or, when the real answers genuinely form a short closed set (a budget band, a yes/no/maybe, picking between a few named directions), a multiple-choice question with 3-6 AI-drafted options; a multiple-choice question also gets "allow more than one" when more than one option can genuinely apply at once (which of these features do you want, which services do you offer), so the client checks boxes instead of being forced to a single pick. If the project sounds like more than a marketing site — a system, app, or anything meant to help run the business day to day — the AI leans into specific, creative multiple-choice questions about real operational features (automated review requests, digital invoicing, instant quotes, scheduling, follow-ups, reporting) grounded in what you actually wrote, not a generic checklist. It never names or alludes to any of your OTHER clients or projects, even as an example — every question stands on its own as if this were the only client you'd ever worked with.

Nothing is saved yet — each proposed question shows with a checkbox, its own editable text, and a Fill in / Multiple choice switch; a multiple-choice question's options are each their own editable line with a remove button, "+ Add option", and an "allow more than one" checkbox. "+ Add my own question" adds a blank fill-in one for you to type yourself — write it (or switch it to Multiple choice) and press "✨ Suggest options" (or "↻ Re-suggest options" to try again) to have the AI propose 3-6 realistic options for that exact question, so you're never stuck typing "Option 1, Option 2" by hand. Uncheck, edit, retype an option, switch types, or delete anything before sending.

"Send N to the client" is the only thing that actually creates the questions — it mints the project's access code if it doesn't have one yet and gives you one link ("Copy link", or "Email it", which opens your own email app with the client's address and the link already filled in; aiKit itself never sends the email). Sending clears the review list and shows a confirmation card with a summary of what went out — the page scrolls down to "Sent so far" so it's clear the questions are saved, not lost, since that same list is where they now live. The client opens that link with no account of their own, picks an option (or checks several, for an "allow more than one" question), or types an answer, and can leave the rest for later. Once they submit, their answers are filed into Brief & rules under Notes so the AI has them on every future edit, and the AI privately checks whether a genuine follow-up is worth asking — most rounds get none, but if an answer is vague or two answers conflict, one to three sharp follow-up questions appear on that same link automatically, no new link needed, capped after a couple of rounds so it can't turn into an endless back-and-forth.

Come back to this page any time — before the client has answered everything, or long after — and send another round with the same Draft-then-review flow; it just joins the list on the same link, nothing new to send. The Sent so far list shows every round, each question with its real answer (and its options, for a multiple-choice one, plus a note when more than one was allowed) or a "Waiting for an answer" pill, refreshing on its own while anything in the newest round is still open. "Export answers" downloads every round's questions and answers as one plain-text file — meant to paste straight into a fresh Claude or Claude Code session to start building, not to open in another tool; the same Q&A also stays on the project's own Brief & rules permanently, so the export is a convenience copy, not the only record. The export also lists every photo the client uploaded, with its note, tag, and link.

Drafts (2026-09-30). A questionnaire can also be loaded as drafts, questions saved on the project but not sent. They appear in their own "Drafts, not sent yet" card with a count, numbered in the order the client will see them. Each draft shows its section heading (if it starts a new section), the question, its translation, its tip, and for a multiple-choice one whether the client picks one or any, plus the choices. Edit opens it in place: the question, the tip under it, and the section heading above it, each in both languages side by side (stacked on a phone); Fill in or Multiple choice; each option in both languages with a remove button, "+ Add option", and "Client can select more than one". The ↑ and ↓ buttons move a draft up or down, the × deletes it (it asks "Delete N?" right in the row, with Delete and Keep), and "+ Add a draft question" adds a new one at the end with the same editor. The client sees none of this. Sending drafts is three steps, so a client link never goes out showing nothing. 1 Review: while drafts are unsent, the top of the Questions page says so in a large orange box ("33 questions not sent yet. David sees nothing until you preview and send.") with one button, "Preview what David will see"; the bottom of the drafts card has "Next: preview". There is no copy-link or email control at this step. 2 Preview: opens the client's own link in the same tab. Because you're signed in to the project, you see the full page exactly as the client will get it (intro, language switch, every question, tips, choices, and the photos and files section) under a bar with only "← Back to edit" and "Looks good, send it". Answers and uploads don't save in preview. Anyone who isn't signed in to the project and opens the link before sending sees only the intro and "Your questions will appear here very soon". 3 Send: "Looks good, send it" publishes every draft at once, in order, and returns you to the Questions page opened at a green panel with the one real survey link, "Copy link", and "Email it to" the client, with the client's email in an editable box (saved to the project) that opens your own email app with a short note and the link filled in; aiKit doesn't send the email itself. That panel stays at the top of the Questions page from then on. Drafts don't count in "Sent so far" until they're sent. No step uses a browser pop-up; every confirmation is inline.

Two languages (2026-09-30). A questionnaire can be written in English and Spanish. The project's primary language is what the client sees first, and each question can carry its translation, including its options, tip, and section heading. When translations exist, the client's page opens in the primary language with an English / Español switch at the top that changes everything on the page at once: every question, choice, tip, section heading, the intro, and all the buttons and messages. The microphone listens in whichever language is showing, so a Spanish speaker can dictate in Spanish. Answers are stored as typed. A multiple-choice answer is always stored in the primary language's wording, whichever language the client was reading, so the Sent so far list and Brief & rules read consistently. A questionnaire with its own intro, its own language, or photos is shown as the business's own page: the header names the project and the aiKit ∞ mark is left off. The intro supports bold text and paragraphs.

Saving answers as they go (2026-09-30). On the client's page every question has its own Save answer button. A saved answer shows a check and the time it was saved, and the client can change it and save again until they finish; anything typed but not saved is kept in their browser, so a refresh doesn't lose it. The progress bar counts saved answers. "I'm done, send everything" at the bottom saves anything still unsaved and hands the answers over: filed into Brief & rules (or held for you under "I review first"), and checked for a follow-up even if some questions were left blank. A round also hands itself over once every question in it has a saved answer. In the Sent so far list, a saved-but-not-finished answer carries a "Saved by the client, not sent yet" pill. A multiple-choice question can also offer "Other" (Add an "Other" choice with a text box, in the draft editor): the client ticks Other and types or dictates their own answer, stored after the picked choices as "Other: …". Follow-up questions on a two-language questionnaire are written by the AI in the primary language with the translation alongside (and a short tip where it helps), based on the client's actual answers and the notes on their photos, at most 8 per round, one follow-up round per round.

Loading a written questionnaire. A questionnaire written as a file (questions in both languages, sections, tips, choices, intro, photo tags) can be loaded as drafts with scripts/import-client-questions.mjs from the aiKit codebase, or through the app at POST /api/hosted-projects/<id>/questions/import (signed-in staff, or a full-access aiKit API key). Either one replaces the project's drafts and refuses if the project already has sent questions; nothing is sent.

Client photos (2026-09-30). The "Client photos" card has a switch, "Ask the client for photos at the end of their questions page." When it's on and questions have been sent, the client's page ends with a Photos section: they pick many photos at once (40 or more is fine), watch each one upload with its own progress bar (three at a time, with Try again on any that fail), then add a note to each (typed or dictated) and tap a tag such as Before, After, Finished job, or Me / my team. Notes and tags save on their own as they type; there's no Send button for photos, uploading doesn't require answering any question, and everything is still there when they come back to the same link. They can remove their own photos. On this page, the card shows every photo with its tag, note, file name, and size, and a Download button. Where a Vercel Blob store is connected to aiKit, photos upload as the original file (up to 25 MB; short MP4 or MOV videos up to 100 MB also work); without one, each photo is shrunk to 2400 pixels on its long side in the client's browser before it uploads, and videos aren't accepted. The section is called "Photos and files": besides phone photos and screenshots (JPEG, PNG, HEIC, WebP), the client can upload GIFs, PDFs, and other files such as a logo, business cards, or flyers (AI, EPS, SVG, Word, Excel, PowerPoint, text, ZIP). Those upload untouched and show as a file card with a Download button; without a Blob store each file can be up to 4 MB. In the staff preview (step 2 above) the section appears exactly as the client gets it, but choosing files only shows a note that nothing is uploaded in preview.

Every change that's actually shipped, newest first, each with a kind pill (fix vs. other), the page it touched, and when. "Copy proof link" copies the same client-shareable, login-free link the Overview tab's "Copy updates link" button does — meant specifically to send a client as visible proof of what's shipped.

This is the one tab with the full plain-language tutorial on domains and where a site lives — written for a non-technical business owner, not a developer. It walks through, in order: where the site lives right now (with a plain headline and a Check-again button, plus a Go-live button if it isn't live yet); what a web address (domain) actually is, explained with a shop-and-street-sign analogy; the difference between the temporary aiKit address every project starts on and the client's own domain; step-by-step instructions for connecting a real domain — the two DNS records to add (an A record to 76.76.21.21 and a CNAME for "www" to cname.vercel-dns.com), plus expandable per-registrar walkthroughs for GoDaddy, Namecheap, Cloudflare, Squarespace/Google Domains, Wix, and Bluehost/HostGator; exactly what changes once the domain is connected (search engines re-index under the real name, an old site on that domain gets replaced, email is untouched as long as MX records are left alone, and the temporary address keeps working); what to check, in order, if the site shows as down; when you'd ever actually need to touch any of this versus everyday Quick edit/Chat changes; and, collapsed by default, the raw technical DNS/HTTP snapshot for a developer or a registrar's own support team.

The plain-language status labels used here — "On their own domain," "On a temporary aiKit address," "Site is down," and the rest — are defined in exactly one place in the code and reused by this tab, the project's own header badge, and the list row's badge, so all three can never say something different about the same project.