# LumenDesk 2.0: what small teams can use now, and what it costs

LumenDesk 2.0 adds a document assistant that answers questions about your files and links each answer to a page. Whether a small team should try it now depends on three things: whether you can get in, what the bill looks like, and how much checking the output leaves behind. The evidence on the third is thin. I have not used the product, and neither has Kingy; everything below comes from the vendor's documents and one short outside review.

## Who can use it today

As of the October 16, 2026 announcement, the assistant is available to existing Team customers in Canada and the United States. New customers can join a waitlist; the announcement gives no date for when they get in. It runs in the browser only [S1-announcement.md, "Access"].

An API and EU availability are planned for Q1 2027. The vendor gives no exact day and does not guarantee the rollout, so treat both as plans, not as something to schedule around [S1-announcement.md, "Access"].

## What it does, and doesn't

You upload searchable PDFs and DOCX files, up to 50 MB each, and ask questions. Answers come with page-linked citations. Scanned PDFs need a separate OCR add-on. The assistant drafts answers; it does not approve payments, send emails or execute contracts [S1-announcement.md, "Features"].

## What it costs

Team is USD 29 per seat per month, billed monthly, with a three-seat minimum. By my arithmetic, that puts the floor at USD 87 a month before tax. The plan includes 500 document uploads per workspace per month, shared across the workspace and not per seat, with no rollover. Extra uploads are USD 0.08 each, and the OCR add-on is USD 12 per workspace per month [S2-pricing.md, "Current plans"].

On those rates, a three-person team that works with scans would pay USD 99 a month before tax, and a month with 700 uploads would add USD 16.

If you saw USD 19 in a launch screenshot, that figure was wrong. The vendor's October 17 pricing update says the screenshot was incorrect and that USD 29 is the current monthly seat price. The pricing document lists no annual offer [S2-pricing.md, "Correction"].

## Where the evidence is weak

The vendor says that in its own internal test of 40 procurement packets, the median time to produce a first draft fell from 50 minutes to 20. The caveats are the vendor's own: the figure covers draft production only, excludes human verification, and does not establish end-to-end time saved. No task-level results are published, so readers cannot check it [S1-announcement.md, "Vendor test"].

The only outside look in the sources is small. DeskLab tried six searchable procurement PDFs. Five answers cited the correct page; one pointed to the wrong page even though the answer text was correct. DeskLab calls this a small, informal review, not a benchmark, and says it did not test scans, OCR, privacy or team workflow [S4-independent-review.md, "Method"].

That one miss matters more than its size suggests. A correct answer with a wrong page reference still sends the person checking it to the wrong place. Six documents cannot tell you how often that happens, only that it can. And checking time is exactly what the vendor's 50-to-20 figure leaves out.

## Where your documents go

The vendor says customer documents are not used to train its base models. They are stored for 30 days by default, and workspace admins can delete them sooner. Deletion from active storage happens within 24 hours, but backups can persist for up to 30 further days [S3-data-policy.md, "Handling"].

Requests are processed by an external model provider under contract, and there is no on-device mode. The policy states that it does not claim zero retention or guaranteed regulatory compliance [S3-data-policy.md, "Processing"]. If your contracts or your regulator restrict where client documents may go, settle that before you upload anything.

## Who might try it now

A reasonable candidate is an existing Team customer in Canada or the United States whose work runs on searchable PDFs or DOCX files, and who already has a person reviewing every answer before anything is paid, sent or signed. For that team, access is open, and the question is whether the assistant reduces the checking.

Wait if you are outside those two countries, are not already on Team, need an API, or cannot send documents to an external processor. Be cautious if you rely mostly on scans: nothing in these sources tests the OCR add-on.

If you do try it, I would start with documents whose answers you already know. Ask the questions, open every cited page, and time the checking as well as the drafting. If the citations hold on your own paperwork, you have a number worth more than the vendor's. If they don't, fluent answers do not make up for it.
