Handsfree

Voice-directed picking for Oracle WMS Cloud, on the devices your pickers already carry.

Hands and eyes stay on the goods. We drive the same RF picking transaction your pickers use today — server to server, into your own Oracle pod. No Oracle customisation, no proprietary terminals, and nothing installed on a single device.

Where this is, honestly: in pilot.

The software is built and tested end to end against a simulated Oracle warehouse. It has not yet run a full production peak, and we have no customer references to show you, because we have no customers yet. Everything on this site is a property of the code you can ask us to demonstrate — there are no testimonials here, no logos, and no productivity figure we have not measured. See what we have and have not measured.

How you would find out if it works

1. Try the demo warehouse — about a minute, no credentials A live warehouse simulation running the real product: the real protocol, the real driver, the real confirmation checks. You sign in as a guest, scan a pallet, walk to a location, say a check digit, confirm an item. It is the whole loop, on your own phone, with nothing to install and nothing to sign.
2. Connect your Oracle QA pod — days, not months Not production. A scoped, RF-only service user on the non-production pod every WMS Cloud customer already has. An eight-step wizard stores the connection, tests it, audits your location labels, captures your menu path and replays a recording of one real pick through the parser. That last step is a go-live gate, not a formality: your pickers should never be the ones to discover a screen we do not recognise.
3. A paid pilot, measured against your own RF baseline Four weeks, up to ten pickers, one facility, $2,500 — credited against your first annual invoice. Paid, because a free pilot never gets an internal sponsor. The deliverable is a number: lines per hour and confirmation-failure rate, voice against the guns, on your floor with your labels and your noise. If it does not beat the guns, you have learnt that for $2,500 and a month.

Prices in full on the pricing page; the subscription starts at $1,089 a month and nothing is quote-only until Enterprise.

What a pick actually sounds like

No screenshots on this page, on purpose — a fabricated dashboard is the oldest trick in enterprise software. Here is the real sequence instead, and you can watch it happen in the demo:

• “Please scan the LPN barcode.” Pallet and staging steps are scans, never voice — a twenty-digit licence plate is a job for the gun, and pretending otherwise is how voice products get abandoned.
• “Go to REG-C-15-10.” — the picker says the check digit Either the code on the shelf label, or, where the labels do not carry one, the last three characters of the rack label already on the shelf. Choosing between those two is a wizard step, not a relabelling project.
• “Confirm the item.” — the picker says its last three digits We always know the expected answer, so the match is against that answer rather than free text. A wrong digit still fails; the prompt repeats and the attempt is counted.
• “Skip”, “short”, “go back” — spoken, and confirmed before they act Exceptions happen dozens of times a shift and a voice product that cannot handle them gets taken off. Destructive answers ask for a spoken yes or no first, and an ambiguous answer counts as no.
• Every voice prompt also takes a typed answer Same handler, same verification, same attempt caps. When the speech vendor is slow, a shift finishes by typing instead of stopping.

Why it installs nothing on a device

The browser never talks to Oracle. Every Oracle call is server-to-server from our backend, which is what makes the rest of this list true:

• Your configuration lives on the server, not on 200 handsets Menu path, verification mode, screen signatures, thresholds — all of it is rows in a database, changed once and in effect on the next prompt. There is no per-device install, no MDM push to schedule, and no fleet half on the old configuration.
• No Oracle credential is ever on a device A lost handset is a lost session, not a lost password.
• We can ask Oracle questions a device-local reader cannot The check digit a picker is asked for is fetched from Oracle's location record, not scraped off a screen — which is also how the wizard can tell you, before you buy, what fraction of your locations even carry one.
• A recorded pick becomes your regression pack Oracle WMS Cloud updates quarterly. Your recording is replayed through the parser in the window between your stage and production waves, so drift is found on a Tuesday by us rather than at 6 a.m. by a picker.

The honest trade is that we are a network service: this needs working Wi-Fi in the aisles and a route out to our backend. A site that cannot keep coverage in its racking should buy an on-device voice product instead of this one. A device that drops out says “You're offline — walk back toward coverage” and reconnects on its own, which covers a dead spot, not a dead network.

Voice is additive. It is never load-bearing.

If any part of this is unavailable — us, the speech vendor, the site's Wi-Fi — your pickers pick up the RF guns and keep working exactly as they did before. Nothing we run sits between your warehouse and Oracle for scanning, and there is no code path from a billing state to a picker's session. That reversibility is the reason a warehouse can take a chance on a small vendor, so we engineer it, contract it, and lead with it rather than promising uptime we cannot staff. What we commit to is graceful degradation; the availability section says so in the same words we would use in a contract.

What version one covers — and what it does not

Voice covers the active-pick zone, not the building. We would rather disqualify you in the first call than discover this in week three of a pilot.

Two more things we raise before a pilot rather than after. This system signs in to Oracle as the picker, so every voice picker needs their own named Oracle account — for some buyers that licence cost changes the decision, and the wizard has a worksheet for it. And headsets are the one piece of hardware that matters: recognition quality comes from the microphone, so a wired or wireless boom mic beats noise-cancelling earbuds every time, whatever the earbuds cost.

What we have measured, and what we have not

• Measured, against a simulated warehouse Roughly 5,000 pick lines a minute before our backend becomes the thing anybody waits for, on a 2019 four-core laptop also running the whole development stack. That is a statement about our code and nothing else: Oracle, speech and the radio network were all deliberately removed from the measurement, and each of them is far slower than we are. It is not a warehouse number and we will not quote it as one.
• Measured, in tests that run on every change Two warehouses picking at once with no cross-talk; a full pick on a renamed, reordered menu driven by configuration alone; disabling an account taking effect on the next request.
• Not measured yet — and only a pilot can Lines per hour against your RF baseline, confirmation-failure rate in your aisles at your noise level, and how a full peak week behaves. There is no number here for those, because we do not have one, and a number we borrowed from somebody else's case study would not be about your floor anyway.

Talk to us