An Aivell runs a manufacturer's paper layer — reads purchase orders arriving by email and enters them in your system, logs into supplier portals for invoices and order confirmations, watches stock levels and drafts reorders, turns shop-floor photos into inspection reports. All on a box inside the plant, next to your ERP.
Customers send purchase orders as email attachments in twenty different layouts, and someone retypes part numbers, quantities and delivery dates into the ERP — the plant's most error-prone workstation.
Order confirmations, delivery notes and invoices live scattered across supplier portals, each with its own password and its own idea of navigation. Someone tours them all, weekly, downloading files by hand.
The reorder point was crossed on Tuesday; the purchase requisition was written on Friday; the line stops in three weeks. Everyone knew — the spreadsheet just didn't tell anyone.
Damaged incoming goods, finished-pallet checks, machine readings — captured as phone photos that die in someone's gallery instead of becoming a report or a claim.
Purchase order, delivery note, invoice — matched by hand, line by line, and mismatches surface only when a supplier calls about an unpaid invoice.
The plant is automated; the office next to it is not. Orders arrive as PDF attachments and get retyped. Supplier confirmations live behind a dozen portal logins. The stock sheet knows the reorder point was crossed, but says nothing. Incoming-goods damage is photographed and forgotten. The three documents that should agree — order, delivery note, invoice — are matched by the person with the least time to do it.
Each step is minutes; together they are headcount. And the retyping is where errors enter: a transposed part number costs more than the hour it took, and teams that automate the reading and entry step report 70–90% less processing time on documents like invoices alone.
Manufacturers have good reasons to distrust cloud tools: pricing, drawings and customer order books are exactly the data competitors would love, and the ERP those tools want to reach is on-premise anyway. An Aivell sits on the same rack. Models run onboard the box; a customer’s PO is read metres from the system it is entered into, and nothing crosses the firewall it doesn’t need to.
The chores read like instructions to a new order-desk clerk. Watch the orders mailbox; check part numbers against the catalog; ask before confirming a strange price; warn planning about rush dates. Tour the supplier portals on Monday and file what’s new. When stock crosses the line, draft the requisition and get the buyer’s yes. When the warehouse photographs a dented crate, turn it into a claim-ready report the same hour.
Guardrails keep it a clerk, not a loose cannon: it appends orders but cannot alter history, works only the portals it is given, and every entry sits in the audit trail. The judgment stays with your people. The retyping doesn’t.
watch orders@ for new purchase orders; ignore quotes and marketing
read the attachment — PDF, scan or photo — and extract customer, part numbers, quantities, dates
check every part number against the catalog database; unknown ones go to Marco, don't guess
price on the PO differs from our list? pause and ask sales
requested delivery under 10 working days? flag the order to planning before confirming
enter the order in the system and file the PO on the share under the customer's folder
reply to the customer confirming receipt and the delivery date, in their language
Purchase orders arriving by email — PDF, scan, photo or plain text — read onboard, checked against your prices, entered into your system, confirmed back.
The Browser Agent logs into supplier, bank and utility portals, downloads the month's invoices and files them on your share — passwords never leave the box.
Stock levels watched nightly in your own database, reorder emails drafted per supplier — and nothing is sent until you approve it with a tap on your phone.
Supplier invoices matched against purchase orders behind your firewall. Clean matches approve themselves for payment; only the mismatches reach a human.
Delivery notes, meter dials and site damage read from photos by a vision model on your own box — and written straight into your database as structured records.
An on-premise AI colleague reads every supplier invoice — PDF, scan or photo — and books it into your sheet or database. Invoices never leave your office.
An on-premise AI colleague watches your database and speaks up only when a threshold breaks — with the numbers, the trend and the context attached.
Usually more easily than a cloud tool can — the box sits on the same network. If the ERP has a database or an import format, the Aivell writes to it directly; if it only has screens, the Browser Agent operates them like a clerk, with guardrails on which transactions it may touch.
The Browser Agent logs in with single-click passkeys, navigates the portal, downloads confirmations, delivery notes and invoices, and files them on your share under your naming scheme. Each portal is a few lines in a chore, and each login is guarded — it can go where the chore says and nowhere else.
On the box, inside the plant. Reading documents, photos and portal pages all happens onboard — your part list, your pricing and your customers' orders never leave the network. It can even run air-gapped from the internet where the chore doesn't need it.
Yes — the vision model runs onboard. A photo of a damaged pallet, a label or a counter becomes structured data in a report, with the photo filed next to it. If the image is too poor to read, it asks for a retake instead of inventing numbers.
Whoever your chore names. Typical rule — draft the purchase requisition automatically, send anything above a threshold to the buyer for a yes from their phone, order routine consumables directly. The threshold is one plain-language line you can change anytime.
An Aivell is an on-premise AI colleague: a small box that plugs into your network, runs its own AI onboard and takes the repetitive work off your desk. Prompts, documents, data — nothing ever leaves your office. Unplug the internet: it keeps working.
You describe each task as a chore, in plain language. Your Aivell turns it into a solid, guarded procedure and runs it in the background — with guardrails on every tool, a complete audit trail, and approvals in your hand when you want them. One-time setup, fixed monthly fee. No tokens, no overages.
onboard inference · air-gapped · fixed monthly fee