- Shipping
- 2-day promise in Sweden and Norway for orders paid before 14:00.
- Returns
- 30 days. Exchanges for another size are free.
- Tone
- Friendly, first names, no exclamation marks.
Run your whole company
from one conversation.
The AI workspace for e-commerce and small teams. Ask anything, build automations that do the work, and teach it your rules once, so everyone gets the same answer.
Early access, by invitation. We are onboarding a few e-commerce companies by hand.
Prepare this week’s business update as a PDF and send me a draft every Monday.
Net sales, week 38
412 300kr
Up 8% on the week before. Sweden and Norway grew, Denmark was flat. Three orders paid on Tuesday have not shipped.
OfficialNet sales, an approved definition · Shopify orders, week 38
- Runs
- Every Monday at 07:00
- Reads
- Shopify
- May create
- One PDF, as a draft for you
- Every system in one place. Shopify, your mailbox, your warehouse and more. The Workshop builds connectors for more.
- Automations that do the work. Describe it once. It runs on a schedule, an email or a webhook, within the mandate you sign.
- Skills the whole team uses. Teach it how a job is done once, and everyone on the team works that way.
- One answer, whoever asks. Approved definitions make net sales the same number for everyone.
Use cases
Your systems.
Working together.
Connect the tools you already run the business on, teach it your rules, then describe what you want done. Pick a team to see how.
Your systems
Company knowledge
What you get
Your workspace
Here’s Alpine’s new price list. Compare it with our price table and flag what needs my decision.
Every morning, check yesterday’s orders against the warehouse and tell me what’s stuck.
Draft replies to new emails in support@ from our returns policy, and hand me the ones it doesn’t cover.
Show last month’s net sales by country as a chart, and put it in a PDF for the board.
alpine-prices-autumn.xlsxPrice table · 14 changes
Before and after · 3 fall below your 40% margin
Preview · you confirm onceStuck orders · 7
Paid yesterday, not shipped · 2 on a 2-day promise
Shopify and Ongoing, checked 07:00Support today
214 in · 187 drafted · 21 to you
Drafted from your returns policyNet sales by country · August
Source and period in the captionTo Alpine Textiles: “Thanks for the new list. We accept 11 of the 14 changes and would like to talk about the other three…”
To the warehouse: “Seven orders from yesterday haven’t shipped. Two are on our 2-day promise, #48213 and #48219…”
To a customer, order #48213: “Your parcel left our warehouse yesterday afternoon. Sorry for the wait…”
To the board: “August report attached. Net sales 1 620 000 kr, up 6% on July; Norway grew fastest…”
When the price table changes
check margins against Shopify · notice in Slack
Within a mandate your admin signedEvery morning, 07:00
compare Shopify and Ongoing · notice in Slack
Within a mandate your admin signedEvery hour, in support@
read what is new · draft within your rules · propose a rule when none fits
New rules go to your wiki editorFirst weekday of the month
the same report · drafted to you at 08:00
Within a mandate your admin signedVerified with a live account Built and reviewed, live-account test pending
Four examples. Build around the way your company works. Illustrative workflows
How it works
From a sentence to a routine.
Two jobs, start to finish, in one conversation. Illustrative examples
- 1AskAnswers with sources
- 2DefineOne answer, whoever asks
- 3TeachA skill the whole team uses
- 4AutomateIt runs on its own
1 · Ask
What were net sales last week, and which products drove it?
Net sales, week 38
412 300kr
Up 8% on the week before.
Source-backed analysis · not official · Shopify orders, week 38
Without an approved definition it still answers, shows how it counted, and says the figure is not official yet.
2 · Define
Make that our definition of net sales: paid orders only, not cancelled.
- Where it comes from
- Shopify orders
- What counts
- Paid orders, not cancelled
- Over what period
- The question decides
- In what currency
- Swedish kronor
- In whose day
- Europe/Stockholm
On Thursday Erik, in purchasing, asks in his own conversation.
Net sales last week?
Net sales, week 38
412 300kr
OfficialNet sales v1 · Shopify orders, week 38
Whoever asks, the official figure is the same, and it says which definition it used.
3 · Teach
When you write our weekly update: net sales first, then the top five products, then anything low on stock. Short. Save that as a skill.
- When to use it
- When it writes the weekly update
- Written from
- This conversation, frozen with it
- Open with net sales, using the official definition.
- The five best-selling products, with change on last week.
- Anything with less than three weeks of stock.
- Five lines at most. No adjectives.
Skills are versioned and approved. Anyone on the team can use this one, in the chat or in an automation.
4 · Automate
Every Monday at 07:00, write the weekly update with that skill and post it in #management.
- Runs
- Every Monday at 07:00
- Reads
- Shopify ordersOOngoing stock
- Follows
- The skill Weekly update v1 and the definition Net sales v1
- May create
- One message in Slack #managementOne per run
- The test run
- 09:31. The update was written, nothing was posted.
From Monday it runs on its own, within what Anna signed. The first posts wait in your review queue until you let it post directly.
1 · Ask
Which autumn products sell slowest but have the most in stock?
More than 12 weeks of stock
5products
OShopify orders, last 30 days · Ongoing stock · read 09:14
Every answer names where it came from.
2 · Plan
Plan a two-week campaign for those five: 20% off, and keep the margin above 40%.
Four of five keep 40% at 20% off. Trail Jacket W would drop to 37%, so it gets 15%.
The table Purchase prices, revision 12 · the skill Campaign checklist
Margins come from your own purchase-price table, not from a guess.
3 · Launch
Create the discount in Shopify and tell the team in #marketing.
- Code
- AUTUMN
- Runs
- 1–14 October
Discounts are not a built-in action, so the Workshop built this one from Shopify’s public API, and Anna approved it after a real test.
- To
- #marketing
Two things leave the workspace, so you confirm twice.
4 · Follow up
Every Monday during the campaign, send me sales for these five against the same weeks last year.
- Runs
- Every Monday at 07:00
- Reads
- Shopify ordersOOngoing stock
- May create
- One PDF in your workspaceOne per run
It works within what Anna signed: read orders and stock, make this report. Anything else, it stops and asks. You pause it when the campaign is over.
Company knowledge
Teach it once.
It follows your rules.
Your people write the rules. Approved knowledge guides its answers and its work. Automations stay within the mandate you approved.
- Rules and policiesThe wiki: a core page it reads every time, searchable pages behind it.
- What your numbers meanDefinitions, approved once. Official figures use that version and say so.
- How the job is doneSkills: the tone of a customer email, the steps of a product launch.
- Lists that live nowhere elseTables for price lists, rates and exceptions, with every change kept.
- Your own notesWhat you write down about how you work. Private to you.
- Written from
- 2 sources in the run
When a case isn’t covered by your rules, the routine proposes one. Your wiki editor decides.
It proposes pages, definitions and skills in the chat, and a person saves them. Anything that steers how it works needs a second person’s approval, or the owner’s.
Data and security
Your data stays yours.
Protections enforced by the database and the platform, not by a policy document.
One company never sees another’s data.
Isolation is enforced in the database itself and covered by the test suite.
row-level security · every company tableConversations are private.
Not even your admin can open a colleague’s conversation. For oversight the audit log shows who asked and the start of the question, never the answer.
chat · readable by its owner onlyPersonal data is masked.
Email addresses, phone numbers, personal ID numbers and card numbers from your systems and files are masked in logs, receipts and what reaches the AI model. Personal data you name reaches the AI only when you expressly allow it. Names and street addresses are masked where a connection points out the field.
a***@***.se · +46 ** *** ** **No background copy of your systems.
Connections are read when you ask, or when a routine you approved runs. Keys sit in a vault, never in the chat.
read 09:14 · key in vaultSupport only when you say so.
Our support can look in only when your owner switches it on, for a set time, and every page they ask to see is logged.
support · 60 min · switched on by ownerNot used for training.
We train no models on your data, and our model providers do not train on data sent through their APIs by default.
models trained on your data · none
Questions
Before you ask.
Something we did not answer? We answer it when we onboard you.
Who is it for?
E-commerce companies and other small teams that run on a handful of systems: a store, a mailbox, a warehouse, a few spreadsheets. You don’t need an IT department.
Do we need a developer to get started?
No. Built-in connectors connect through a secure form or the vendor’s own sign-in page. For a system without one, the Workshop builds a connector from the vendor’s public API, and your admin approves it after a real test.
Which systems does it work with?
Built in today: Shopify, Google (Gmail, Calendar and Drive), a company mailbox on Google Workspace, Fortnox, Ongoing WMS, Slack and GitHub, plus notices to Microsoft Teams. Shopify and Google are verified with live accounts; the others are built and reviewed, and are tested with live accounts as companies join. Other systems with a public API can be connected through the Workshop when they sign in with an API key or a token; sign-in with OAuth for a system you connect yourself is not built yet.
What does it do without asking?
It reads, researches and prepares. Anything that leaves the company, an email, a message or a change in another system, waits for a person to approve exactly that, or runs within a mandate your admin has signed.
What happens when it isn’t sure?
It says so. If a check cannot be completed, it tells you what it could not verify instead of guessing, and an unclear outcome is never retried on its own.
What can it produce?
Answers with sources, charts from your own numbers, and files: PDF, Word, Excel, PowerPoint and CSV. It also reads the PDFs, spreadsheets, documents and images you upload.
What does it cost?
We publish pricing before launch. During early access we onboard a few companies by hand, by invitation.
Start with one task in your business.
During early access we set up one weekly task with you, and you decide what it may do on its own.
There is no open sign-up yet. Invited companies get their account from us, and colleagues are invited by their admin.