Choose how deep a research request goes

Goal

New in release 26.37

Live since September 10, 2026. The depth picker shipped with this release. (Corrected in 26.39: the card shows a time and search estimate per option, not a token estimate.)

Decide, per request, how much research Maggie does — and therefore how many Intelligence Tokens it uses — before the work starts.

Before you start

• Research draws on your organization's Intelligence Token balance. The deeper the research, the more it uses.

Steps

1. Ask for research in conversation the way you normally would: "Research Acme Robotics before Thursday's call" or "Build me a dossier on this supplier."

1. If the subject could mean more than one thing, Maggie shows the candidates she found. Pick the right one; everything after this runs against your pick.
2. Maggie asks how deep to go and shows a card with three options:
   • Quick — highlights only. A few targeted searches covering the core facts, the commercial angle and matters of record — Quinn, Phoenix and Avery. Right when you need a briefing in minutes.
   • Standard — the default. Every research angle, one search each.
   • Deep — every angle, several searches each, with cross-checking. Right for due diligence or anything you will put your name to.
3. Each option on the card carries its estimate — how long it takes, how many web searches it runs, and which companions run it: Quick (2-3 min) · ~3 searches · Quinn, Phoenix, Avery, Standard (5-10 min) · ~6 searches · all 6 companions, Deep (15-20 min) · ~18 searches + a deep-search tree (up to 100) · all 6 companions. Quick is three of the six companions, and the card names them, so you can see what a quick run leaves out before you pick it rather than after. The search counts and the companion names both come from the same plan the research runs, so what the card quotes is what runs. Deeper research uses more Intelligence Tokens.
4. Tap an option. The research starts, and progress appears in Tasks and in the conversation.

If the card cannot be shown (for example, you asked from a channel that cannot display it), the request runs at Standard.

Reading the result

• The finished report links to the sources it used. Where a section rests on no source at all, the report says so instead of presenting a guess as a finding.
• Share a finished dossier from its page with Share Dossier, by email address or by link.

Verify it worked

Open Billing → Usage after the report lands. The research appears there with the tokens it used — more for Deep than for Quick.

Troubleshooting

Problem	What to do
I was not asked for a depth	Depth was already implied by your wording (for example "quick check on…"), or the channel could not display the card; it ran at Standard.
Deep seems like a lot	Choose Quick first to confirm you have the right subject, then run Deep on that subject.
The card says live web search is degraded	Sources will come from AI-grounded search instead, with fewer cited pages. You can continue, or wait and ask again later.
The report has few or no sources	Some subjects have little public record. The report labels these sections rather than padding them; try widening the subject or adding what you know.

Related

• Understanding Intelligence Tokens
• Talking to Maggie