Skip to main content

Using the Assistant (Beta)

Written by Taylor Stewart

The Assistant is an in-app AI helper that lives next to each client's plan. Open it from the Assistant tab in the client sidebar (Sparkles icon, marked Beta) and ask it anything about that client. It can summarize the plan, surface gaps, draft client emails, look up tickers, and propose changes to the data — all scoped to the client you're currently in.

It's in beta. The shape will keep changing.

Where to find it

Open any client and look in the left sidebar. Assistant is the bottom item, below Office. Only advisors see it — clients in the portal don't.

Each client has its own Assistant. Threads belong to that client; switching clients switches threads.

What you can ask it

The empty state shows four starter prompts:

  • Summarize this client's plan

  • What's their retirement gap?

  • Draft an email about the latest changes

  • Review their goal allocations

Those are starting points. Anything that lives in the client's plan is fair game — assets, liabilities, income, goals, holdings, tasks, entities, documents. You can also ask it to look up a ticker by name or symbol and it'll pull live price, fundamentals, sector mix, and a chart.

How it works

The Assistant has read access to the full client store and can write JavaScript against a frozen snapshot of it to answer questions. When you ask it to change something — add an asset, update a goal assumption, mark a task complete — it doesn't write directly. It drafts a proposal, you see the diff in an approval panel, and nothing persists until you click approve.

That means:

  • Read-only questions ("what's their savings rate?", "show me top holdings") return an answer in the chat.

  • Change requests ("add a 401k for John worth 80k", "set the retirement goal to age 67") show up as a proposal with an approve/reject button. Reject and nothing happens.

You can attach files (paste an image, upload a PDF) and the Assistant will read them as part of the question.

Threads

Each conversation is its own thread. You can have many per client. The URL updates as you switch — /client/<id>/agent/<thread-id> — so threads are bookmarkable and shareable inside your team.

Start a new thread by going back to the Assistant tab from the sidebar (the empty state with the starter prompts is the "new thread" view).

What it can't do

  • Reach across clients. Each Assistant is scoped to the client you're in. It can't compare two clients or pull data from another file.

  • Touch anything outside the client store. No CRM, calendar, email send, or third-party API access.

  • Make changes without your approval. Every write is a proposal.

Auto-triggered runs after document extraction

When the Extractor finishes processing a batch of documents for a client, the Assistant kicks off a thread automatically to review what was found and flag anything worth your attention. You'll see the new thread in that client's Assistant tab — open it to see the review and approve or ignore any proposed follow-ups.

If you'd rather skip the auto-review, just close the tab. It doesn't change anything until you act on a proposal.

Tips

  • Be specific. "What's their retirement gap assuming a 4% withdrawal rate?" beats "Are they on track?"

  • Approve carefully. The diff shows every event the Assistant wants to apply. Read it before clicking approve, especially for batch updates.

  • Use it for first drafts. Drafting a client email, summarizing the plan, or reviewing allocations is where it shines. Treat the output as a starting point, not a finished deliverable.

Known limits in beta

  • Long, multi-step changes can take a few rounds of back-and-forth to land cleanly.

  • It occasionally misses context that's only in a document until it loads that document explicitly. If an answer feels off, ask it to re-check using the document.

  • Output formatting in the chat will keep evolving.

If something looks wrong or the Assistant proposes a change you didn't ask for, reject the proposal and message support — the trace helps us improve it.

Did this answer your question?