The proposal assistant

Updated July 11, 2026

Preview note

Preview note: in this release, the assistant thread handles a small, specific set of commands for real. Setting the agency fee or the policy fee genuinely edits the deck and recalculates the total. Adding the umbrella or the flood recommendation genuinely adds it to the Recommendations slide. Every other request returns a scripted preview response describing what the full version will do, it does not yet touch your proposal. The full conversational assistant, wired to the Anthropic API against your Vault and this proposal as context, arrives with the hosted release.

#Where to find it

Open Outline on any built proposal. The Assistant thread sits at the top of the left pane, with an input placeholder that hints at what to try: "Tell it what to change, e.g. set the agency fee to $650, add the umbrella." Type a request and click Send.

#What genuinely works today

Three requests have real effect on the proposal, not just a reply in the thread.

Setting a fee. Type something like "set the agency fee to $650" or "set the policy fee to $300." The assistant finds the matching fee line on the Premium summary, overwrites the amount, and recalculates the total, exactly as if you had typed the new amount into the outline field yourself. It writes back a confirmation naming the new fee and the new total. If the proposal has no premium yet, for example a discovery brief with no carrier quote, it tells you plainly there is nothing to stamp yet. If it cannot find a matching fee line at all, it says so instead of guessing. See Edit fees and totals for the other two places you can make this same edit.

Adding the umbrella recommendation. Type something like "add the umbrella." If the coverage gap scan flagged an umbrella or excess-liability gap on this proposal, the assistant adds it to the Recommendations slide and confirms with the specifics it found on the quote.

Adding the flood recommendation. Same pattern, for a flood-coverage gap. Type "quote flood" and, if the gap scan flagged one, it lands on the Recommendations slide with a note tying it back to your Vault's coverage standard.

#What is still a preview

A few request types are deliberately refused rather than mocked. Ask for a deductible or "AOP" change, for example, and the assistant will not silently alter it, a deductible change needs the carrier to re-quote. It tells you to edit the coverage table directly if you want the displayed value changed, and flags it as something to raise in the binding conversation.

Anything outside the handful above returns the same honest fallback: in production this goes to Claude with your Vault and this proposal as context, and the edit lands instantly. For now, it points you back to the commands that do work, setting the agency fee, adding the umbrella, or quoting flood.

#Working alongside the canvas and outline

The assistant is one more way to reach the same fee edit you can also make directly on the canvas or in the outline fields, not a separate system. Whichever way you make a fee change, canvas, outline field, or assistant command, the total and the side-by-side comparison stay in sync. See Edit slides on the canvas and The outline editor for the other two paths.