AI & Automation

SKU.io puts AI where it removes real work: reading the invoice a supplier emailed so nobody re-keys it, turning "orders shipped last month over $500" into a filter, drafting a reorder quantity you can audit line by line, and answering "where is this order?" without you learning another screen.

There's no single AI mode to switch on. Each piece sits inside the job it belongs to, and all of them run on your AI provider account — you choose the provider, the model, and who can use it.

At a glance

What it does	Where you'll find it
Look things up, create records, and make changes by asking	The AI Assistant, from the top of any screen
Turn a sentence into a filter on any list	AI Filter in the toolbar above any table
Build a report by describing it	Insights → Custom Reports
Redesign a printed document by describing the change	Settings → PDF Templates
Read supplier invoices, customer POs and credits into records	Orders → Document Inbox
Suggest reorder quantities, and show how each was reached	Inventory → Demand Planning
Run repetitive jobs on a trigger or a schedule	Workflows
Route out-of-stock orders to a drop-ship supplier	Supplier and warehouse settings
Let your own AI agent drive SKU.io	Agentic Skills and the assistant connection

Ask for it in plain English

The AI Assistant

The AI Assistant opens from the top of any screen, either beside your work or as a full page. It's grounded in your account's real data, and it can act, not just answer:

• Look things up — stock on hand, the status of an order, a customer's history, an overdue purchase order, margin on a line.
• Create and change records — a sales order, a stock adjustment, a warehouse transfer, a supplier.
• Stay on topic — focus a conversation on Products, Orders, Inventory, Customers, Purchasing or Returns, or leave it on General.
• Work from a specific record — type @ to mention an order, customer or SKU and pull it into the conversation.

Conversations are saved down the left, so you can pick one back up or run two side by side.

Nothing is written until you confirm it

When the assistant is ready to create or change something, it stops and shows a Confirm card first: exactly what it's about to write, which record it changes, and the permissions involved. Nothing reaches your data until you click Confirm — and the assistant can't delete records at all, or change users, permissions, settings or integration credentials. See What the assistant is allowed to do.

Filters on any list

Every list in SKU.io has an AI Filter button in its toolbar. Describe what you want to see — "all orders submitted in the last 30 days where shipment status is pending and tax is not included" — and it becomes the table's own filters, which you can then read, adjust and save as a view like any other.

Because it lands in the normal filter builder, you can check what it did before you trust the numbers, and save it as a pinned view once it's right.

Reports

In the custom report builder, Report Builder AI takes the same approach: describe the report you need and it assembles the columns, filters and grouping, which you then refine by hand.

Printed documents

The PDF template editor has an assistant that rewrites your packing slips, invoices and purchase orders from an instruction like "make the items table full width" — with a side-by-side diff of what changed, and Undo if you don't want it. It knows which fields exist for the document type you have open, so it can't invent a placeholder that won't print.

→ Edit a template with the AI assistant

Documents that read themselves

The Document Inbox is one queue for the business documents that arrive as files rather than orders you type in. Customers and suppliers email a PDF to your account's intake address — or you upload one — and SKU.io scans it, pulls out the header and line detail, matches the contact and each line to your own records, and holds it for review.

Four document types become four records: a customer PO becomes a sales order, a supplier invoice becomes a purchase invoice, a vendor credit becomes a vendor credit, and a landed cost invoice becomes landed cost bills.

Nothing posts on its own. SKU.io does the reading and flags where it's unsure; you approve, and the record is created.

→ How the Document Inbox works

Forecasts that show their working

Demand planning is the one place where the answer is statistics rather than a language model — deliberately, because a reorder quantity has to be defensible.

• Trend-aware baselines follow a product's actual growth or decline instead of a flat average, damped so a short spike can't run away with the order.
• Anomaly detection keeps one-off spikes out of your everyday demand.
• Confidence rides on every line — High, Medium, Low or New — with a likely range, the drivers behind the score, and a suggested fix when it's low.
• Promotion lift is measured from your own history and stored as a multiplier, so a sale that lifted demand 8× still lifts it 8× after the product has grown.
• Every number is auditable — open a line to see base velocity, the trend factor and the final baseline, and export the whole calculation to CSV or Excel.

→ Forecast with trend-aware baselines · Create POs from a forecast

Automation that runs without anyone watching

Not everything worth automating needs a model. Workflows is a visual builder for the repetitive jobs you'd otherwise do by hand: import products from a spreadsheet, sync a supplier feed, pull orders from a partner, route each order to the right warehouse, or email you a reminder on a schedule. Start from a template or build your own on the canvas.

→ Workflows · Create a workflow from a template · Route orders to the right warehouse

Drop-ship automation is the same idea applied to fulfillment. An out-of-stock order for an eligible product routes to your drop-ship supplier instead of going on backorder, and the supplier purchase order is created and sent for you — immediately, or batched on a schedule you set. A status panel spells out exactly why an order routes to a supplier, or what's missing when it doesn't.

→ Amazon and drop-ship POs · Create a drop-ship request

Bring your own AI agents

Everything so far is AI inside SKU.io. This section is the reverse: your own AI agent, outside SKU.io, driving your account.

Agentic Skills are open source, curated shortcuts that tell an agent exactly which SKU.io endpoints to call for a real business task — build a product catalog, create a sales order, adjust inventory — so it doesn't have to rediscover the API. Each skill is authored once and compiled for Claude, OpenAI, and Gemini. They run in a terminal or an agent platform against your account's API, which makes them the right tool for bulk and scripted work rather than a conversation.

→ Agentic Skills · the open source library

Connecting an assistant covers the interactive side. SKU.io speaks the Model Context Protocol (MCP), so Claude Desktop, Cursor, Claude Code, or any other MCP client can read your orders, stock, and suppliers, and act on them directly. You generate one connection token, paste a config, and the assistant acts as you — with exactly your permissions, nothing more.

→ Connect an AI assistant

How AI is set up — and what it's allowed to do

You bring the provider key

The AI features in SKU.io run on your own account with an AI provider, configured once under Settings → AI Assistant:

• Choose the provider — Anthropic, OpenAI, Google, or DeepSeek — and paste its API key. The key is encrypted at rest, and Test connection confirms it works.
• Pick the default model for the workspace. Anyone can switch model for a single conversation, and a stronger model gives more accurate results.
• Set a rate limit — requests per hour, per person.
• Add global instructions that apply to every conversation, for house style or standing context.
• Watch usage — conversations, messages and tokens — and read the log of what ran.

Until a key is added, the AI features stay off, and each one tells you where to turn it on. Table filters can also be switched off for the whole account on their own, independently of the assistant.

What the assistant is allowed to do

• It acts as you. It sees and changes exactly what your own permissions allow — it has no role of its own.
• Writes are confirmed, one at a time. Creating or changing a record always goes through the Confirm card described earlier. This is enforced in the product, not a convention the model is asked to follow: the assistant has no way to send a write without producing a card for you to approve.
• It cannot delete. Deletion isn't available to the assistant at all — do those by hand.
• Sensitive areas are off-limits. Sign-in, users, permissions, settings, webhooks and integration credentials can't be changed through the assistant, whatever it's asked.
• AI answers can be wrong. The assistant reads your live data and says where a number came from, but check anything you're about to act on — the same as you would a colleague's spreadsheet.

Where to start

• Curious what it feels like? Open the AI Assistant from the top bar and ask it about an order you already know the answer to.
• Drowning in supplier paperwork? Start with the Document Inbox.
• Buying too much or too little? Start with trend-aware forecasting.
• Doing the same thing every week? Start with Workflows.
• Building on the API? Start with Agentic Skills.