Draft with AI through the gate
The drafting gate, one section at a time, and how the AI chat changes your text.
In short
Drafting is blocked while any assumption is still pending, and the AI drafts one section at a time in dependency order. Every sentence it writes must come from your brief, your data or a paper in your vault, and you approve each section before the next.
1Clear the gate
The checklist shows pending assumptions. Confirm or edit each one, or record a waiver with a reason. The gate re-runs whenever a new assumption appears mid-draft.
2Draft the next section
Draft <section> runs in dependency order: Methodology → Results → Related Work → Introduction → Conclusion → Abstract → Title. You see the section as a diff, apply it, then approve it (or edit it first).
3Use the AI chat for anything else
The floating chat icon opens a resizable panel. Ask for a rewording or a fix; changes come back as edits over the exact current text, applied only when you press Apply.
4Sign off figures one at a time
A drafted figure is inserted at natural size with its label and caption and waits for your sign-off before the next one. Generated is not done.
For technical users
Draft calls go through the step gateway (announced steps, 90 s still-waiting, Stop, visible retries, switch-and-retry). Every sentence comes back with a trace; an assumption the model had to take becomes a pending register row that re-runs the gate. The chat may edit the open file in place in a CLI agent's workspace; merge-back becomes the same range diff. A drafted citation of a vault paper with no bib entry adds @misc from the vault's own title only. An imported single-file paper cannot be drafted section by section — the app says so.