Skip to main content

The Assistant

The assistant is a single chat sidebar available on every page of the app. Open it with the chat button or ⌘L (Ctrl+L). It replaces the per-page chat panels older versions had: one conversation surface that knows where you are and what you are looking at.

Page awareness​

The assistant receives context for the page you have open. In the content editor it sees the draft; on a calendar card it sees the card; on engine pages it sees the section you are editing; on analytics tabs it sees the data behind the view. "Make this shorter" or "why did impressions drop this week" work without explaining what "this" is.

What it can read​

Beyond the current page, the assistant has read tools across the workspace: analytics (overview and per-platform), the content library, individual pieces, the engine, insights, and the schedule of upcoming posts. Two are worth knowing by name: search_library searches everything the workspace has made by meaning (pieces, docs, learned memories, competitor profiles), and get_workspace_status answers plan, usage, quota, and connection-health questions with real numbers. It never estimates workspace facts from memory; it fetches them. It can also run live web research for anything outside the workspace, like competitor pricing or current events.

Attach context with @​

Type @ in the composer to attach specific items to your message: content pieces, reference docs, competitors, and campaigns. An autocomplete opens as you type a name; queries of three or more characters also search the library by meaning, so @pricing can surface the piece about costs even when its title never says "pricing". Arrow keys move the selection, Enter or a click inserts, Esc closes.

An inserted mention sits in your text as a plain token:

Compare @[Q3 launch recap](content:abc123) with what @[Acme](competitor:def456) is doing

When you send, the assistant receives each mentioned item in full alongside your message (up to six per message): the piece or doc body, the competitor profile, or the campaign plan. Mentions beat page-context guessing when the thing you're asking about isn't the page you're on. Square brackets can't appear inside a label; picking an item with brackets in its name replaces them with spaces.

Review before apply​

The assistant never writes directly. Anything it wants to change arrives as a card in the chat, and nothing lands until you act on it:

  • Revisions: proposed rewrites of the piece you have open. Review changes shows the edit as an inline diff in the editor; accept or discard it.
  • Proposals: new engine entries (audiences, landscape items, context). Select the ones you want and apply.
  • Field updates: edits to engine fields (voice, brand, positioning, rules, platform tone). Applying fills the page form; you still save the page.
  • Scheduling: a proposed calendar slot for an existing piece, with the date shown in your local time. Applying books it; the server re-checks ownership and status.
  • Guard rules: state a durable preference ("never call it a platform", "remember: sign posts with the team name") and the assistant proposes a standing rule. Applying adds it to the Anti-Slop Guard as a banned-phrase lint or a skill.
  • Campaign plans: multi-piece plans arrive as a campaign card with the goal and piece list; Save keeps a draft, Build creates the briefs and calendar cards. See Campaigns.

It can also render charts inline when a question is best answered with one.

Queue follow-ups​

Sending while a reply is still streaming doesn't interrupt it. The message queues instead (up to three), shown as dimmed bubbles under the composer, and each queued message auto-sends in order when the current reply finishes. A failed reply still dispatches the queue. Remove a queued message with the ✕ on its bubble; the stop button still aborts the current reply.

Model picker​

The assistant uses your selected model (the picker sits in the chat dock, grouped Economy / Standard / Frontier). Usage draws from your AI allowance like any other generation.

History​

Navigating to another page parks the current conversation in History and starts fresh with suggestions for the new page. Reopen any parked chat from the history list to pick up where you left off.