Doc

You don't write pages; you ask for them ("my tasks this week", "the Q3 pipeline by stage"). The assistant calls renderView, creates a draft page, and streams a deep-link pill into the chat. Open it, then keep refining from the chat dock without leaving the page.

Every data block is bound to a live workspace primitive (tasks, contacts, files, deals, entities, memories), not a snapshot. Open a saved page tomorrow and it re-resolves against current state; edit the primitive and the page follows. The page is a view, not a copy.

Every renderView starts as a draft. Drafts auto-prune 30 days after their last touch (any read or edit resets the clock); click "Save" and the page lives forever, click "Unsave" to drop it back to draft. The sidebar lists both groups side by side.

Pages are composed of blocks: text, heading, divider, data, chart. Data blocks render as Table, Board, KPI, or bar / line / pie charts through the same share-safe renderer chat uses; no model output reaches the DOM directly, and everything is dark-mode aware.

Doc editing is a skill, not a dedicated assistant. Whichever assistant you talk to on the Doc surface gets the page tools automatically and is steered to render rather than narrate: "show me" requests start with a view, analytical questions still answer in prose.