Ask Viewfield

Ask Viewfield is the one chat in the studio. There is no chat panel beside your work: it lives behind one box that opens from anywhere.

Opening it

• ⌘K at the foot of the rail (it reads Ctrl K off a Mac), or the keys ⌘K or ⌘J (Ctrl K, Ctrl J).
• Ask in the top bar on a phone or any touch screen.
• An old link to /sessions opens it too.

The box says "Go anywhere, make anything, or ask". Type a question and the first row is Ask Viewfield with your words; the other rows create a piece ("Create a blog post", an X post, a LinkedIn post, a newsletter, or "Just chat"), go to any place or page by name, find drafts and AI visibility questions by their words, switch workspace, and hold your past chats (the three newest before you type, any of them by the words in its title). Asking opens the conversation as a page of its own on Create, carrying the page you asked from, so a question about the draft you had open is about that draft; a past chat resumes where it left off.

It is where you research before you write, think a draft through, ask for a change, teach the brand a rule, or read what AI answers say about you.

What it reads

The chat reads across the workspace with the same tools an API key has, so its answer matches the page:

• Your drafts: the library by title, kind and state, and any draft's text, found by meaning with search_library.
• Brand: the voice, the rules, the rule for each kind, the checks, your docs.
• AI visibility: how often AI answers name you, each question's answers, the sources and pages they cite, one answer word for word, a measured change with its counts, and the next steps.
• Your website: the pages a connected CMS reaches, one page read now, what the site serves on a question.
• The live web: web_research runs live searches and returns a sourced summary (one call can cover several subjects); read_url reads a page you paste. These are the most expensive thing in a conversation, so it asks a clarifying question first and researches once, against a settled ask.
• The workspace: your plan, the allowance left, your notifications.

It never estimates a workspace fact from memory: it fetches it.

What it can do, area by area

Area	It can	It cannot (a person's press in the app)
Drafts	Draft a piece; plan up to three pieces for a tracked question; list, read and search drafts; rewrite a draft; mark it Final or Draft; change its kind; see and restore versions; check its claims; delete a draft (manager); prepare a CMS preview and read where a draft stands in your CMS	Save a draft in the CMS or publish it
Research	Search the live web, read a page, read the workspace's plan and usage
AI visibility	Read visibility, questions, citations, answers and changes; add, reword, pause or archive questions and change the settings (manager); read the next steps; read your pages and what the site serves on a question; check a page again and choose the Website AEO site (manager); propose a page change and read proposals	Save a page change in the CMS, publish it or revert it; run an AI answer check now (the schedule's alone)
Brand	Read every section; update the description, perspective, voice, tone and writing rules (manager); set a kind's rule and delete a custom kind (manager); switch checks and manage learned rules; add, edit or remove a doc (manager)	Add or remove a company on the list of who else gets named (the Visibility board, or MCP); change the website (it re-reads the site)
Settings	Read and clear your notifications	Billing, API keys, members and roles, connecting a CMS or Google, deleting or transferring a workspace

AI visibility changes happen when you ask

Ask Viewfield to change what AI visibility tracks and it does it, in the same reply: "refresh our questions for what we sell now" reads your questions, your plan's limit and who the answers name, then adds the competitors, archives the questions that no longer fit, adds the new ones (a whole set at once, under the categories it names) and sets the buyer category. Each step leaves a Done card listing what changed, and the Visibility board shows it straight away. New questions are asked on your plan's engines within minutes, not at the next weekly check. Your role is checked on the server for every change, so the chat cannot do what you could not do yourself. One change still asks first: the names and the sites that count as your brand, because every write to your site is bounded by them.

Every other change asks first

Nothing else changes because the chat decided to. A tool that writes is a card the chat shows before it runs, saying what it is about to do, with Confirm; your role is checked again on the server the moment you press, so the chat cannot do what you could not do yourself, and a manager-only change is refused before a contributor ever sees its card.

You ask it to	The card, once you confirm	Who can
Write something	A draft is written (draft_piece)	Any member
Rewrite a draft	The revision comes back as a card: Put it in the draft when that draft is open (it lands in the editor and asks Keep or Revert), or Open the draft	Anyone who can edit it
Never say a word again, always follow a rule, sound warmer	The rules change and every draft from then on obeys (set_writing_directives)	Manager
Change the names and sites that count as your brand	The AI visibility settings change (manage_aeo_tracking); every other AI visibility change happens at once, above	Manager
Plan content for a question	Up to three pieces, each with its form, working title and argument, saved on the question (plan_content)	Any member
Propose a change to one of your pages	A few changes with their before and after, saved as a proposal on the page; nothing on your site changes (propose_site_change)	Any member
Prepare a draft for your website	A preview of the draft as it would be sent to a destination; nothing is sent (prepare_site_publication)	Any member

Files, follow-ups and stops

The conversation's own box takes files with its attach control: up to five, 10 MB each (text, Markdown, CSV, PDF, Word, Excel or Numbers), whose text is read with your message. Sending while a reply is streaming queues the message (up to three) instead of interrupting. Stop halts the reply; a tool already running finishes. A reply takes at most fourteen steps, offers at most three cards, makes at most twelve AI visibility changes and creates at most three things, so a runaway ask cannot spend a month's allowance; when it runs out of steps it says so and you can ask it to carry on. Chats are private to you; drafts are shared.

Over MCP

send_chat_message and get_conversation drive the chat on one draft. Drafting is draft_piece. Nothing over MCP saves to a CMS or publishes. See MCP overview.