It is the system a business runs on. Not a CRM with modules bolted to it — one record for the whole operation, and an AI back office that reads it and answers in plain language.
Almost nobody chooses to run their company across six disconnected tools. It happens one sensible decision at a time.
You start with a spreadsheet, because a spreadsheet is free and you have eleven customers. Then you get a webshop, because that is where the customers are. Then an accountant asks for something that looks like books, so you buy accounting software. Then you hire someone, so there is a folder of passport copies and visa expiry dates. Then you want to email your customers, so there is a mailing tool with its own separate list. Every one of those was the right call on the day.
The cost arrives later, and it arrives as disagreement. The shop says one revenue figure and the books say another. The mailing list has people who stopped buying two years ago and is missing everyone who bought last month. Nobody can say what the best-selling product actually earns once the discounting comes out, because the sales sit in one system and the costs in another. The business is not badly run. It is simply not in one place, and so nobody can see it whole.
The phrase is deliberate. An operating system is not an application — it is the layer everything else runs on, that owns the files, the memory and the permissions, so the applications above it do not each have to invent their own.
OperateOS does the same for a company. It owns the record: the customers, the documents, the stock, the ledger, the people. Every function reads and writes that one record rather than keeping a private copy. When a sale happens, the customer, the stock movement, the ledger entry and the tax consequence are the same event seen from four angles, not four separate entries someone has to reconcile later.
That is the whole design. Everything else — the AI, the reports, the automation — is only possible because there is one record underneath to reason about.
A CRM is one of the eleven. That is the argument in a sentence.
Every buyer, their whole order history, the companies behind them and the people inside those companies. Trade accounts and individual consumers in the same place, because a business that sells to both should not need two systems to see it.
Contracts, invoices, statements, certificates. Every document keeps its original and a SHA-256 hash, and the text extracted from it is stored alongside rather than instead. A derived value never replaces an evidenced one — if the reading is wrong you can always go back to the paper.
Photograph a supplier invoice with your phone. It is read, matched to the supplier, coded to an account and presented as a bill waiting for approval. Nobody types it in, and nobody approves it by accident — it sits until a person says yes.
Stock you hold, and stock sitting at a retailer that is still yours until it sells. Consignment is where most small systems give up, and it is exactly where the money hides for a business that sells through other people's shelves.
A real ledger with debits and credits, not a category column on a list of payments. It matters because only a real ledger can be checked: a trial balance either balances or it does not, and a spreadsheet always balances because nothing is stopping it.
The return prepared from your own transactions, with every figure traceable back to the invoice that produced it, and the supporting schedule assembled with it. The rates, the periods and the treatment of exempt and zero-rated supply are configured per country, so the same ledger serves whichever regime you file under. A person files it. The platform prepares and evidences; it does not submit.
Business email and campaigns, with consent as a first-class record rather than an assumption. Consent is checked again at the moment of sending, and contacts without it are withheld and reported rather than quietly included. A purchase is not permission.
Visas and work permits, trade licences, tenancy contracts, insurance, certifications — everything with a date on it, counted down before it lapses. This is not administrative tidiness: in most jurisdictions an expired licence stops the business trading and an expired permit stops a person working, on the day, with no grace period.
Close a month and get a profit and loss, a balance sheet and a cash position where every figure can be opened down to the transaction that made it. No number appears that cannot be explained.
Thirty defined procedures run over your own figures — populations drawn from the source and recomputed independently, never accepting a stated total. Exceptions are quantified and any procedure that could not be performed is reported as a scope limitation rather than dropped. It expresses no audit opinion. We are not a licensed audit firm and the report refuses to pretend otherwise.
Ask your business a question in plain language and get an answer built from your own records. The agent reads; it never writes. More on that below, because it is the part that deserves the most scrutiny.
Twelve desks, each of which knows one part of the business and answers only from your data.
Nine of them are in-house: sales, finance, inventory, marketing, compliance, management, operations, people and assurance. Ask the finance desk what the month closed at and it reads the ledger. Ask the inventory desk what is not moving and it reads the stock record. The answer arrives with the figures it used, so you can check it rather than trust it.
Three more are different. The credit, advisory and capital desks read a professional practice's own method at the moment you ask — the same method that firm uses on its own engagements — and apply it to your numbers. That is the difference between a competent general answer and a firm's answer.
What none of them do is guess. A desk that has not been given the facts to answer a question says so and names what is missing. An answer invented to be helpful is worse than no answer, because you cannot tell the difference until you have acted on it.
An AI with silent write access to a company's ledger is not a feature. It is an unresolved risk with a friendly interface.
Four rules, held in the architecture rather than written in a prompt — the distinction matters, because a prompt is a request and architecture is a wall.
A CRM knows your customers and nothing about your money. It will tell you who you spoke to and never what they are worth once costs are taken out.
Accounting software knows your money and almost nothing about your customers. It can tell you revenue was up and not which customers left.
A traditional ERP does join these up, and that is why implementations run for a year, cost more than the software and usually need a consultant permanently attached. They are built for companies with a project team to spare.
OperateOS is the middle that is missing: one record like an ERP, set up in days rather than quarters, and an AI layer that removes most of the reason you needed the consultant — because you can ask the system what you want to know instead of learning where it is buried.
Most software treats tax as a percentage and compliance as a reminder. In practice a return has a defined shape, the evidence behind it has to be retained and produced on request, and a lapsed licence or work permit does not generate a warning — it stops you trading and it stops your people working. That is true in Dubai, in Riyadh, in London and in Singapore; only the forms and the dates differ.
So expiry compliance is a module rather than a calendar entry, tax is built into the ledger rather than calculated at the end, and the base currency is whatever your business actually keeps its books in — with proper handling for the many that trade in several.
You connect what you already have — the shop, the bank statements, the invoice history, the spreadsheet of stock. Nothing is switched off, nothing is migrated over a weekend, and your existing systems keep running for as long as you want them to.
Then you read what it found. This is consistently the part that changes people's minds, and it is rarely comfortable: customers nobody had a record of, orders nobody counted, stock that has not moved in a year. On one live business in a single afternoon it surfaced 1,460 orders the company had never seen, 569 buyers with no customer record, and a card-testing operation of 321 fake customers that had pushed the cancellation rate to 74% and had been running for months.
Then you ask it something hard, and decide for yourself whether the answer is any good.
It does what an ERP does — one record across the whole operation — without the year-long implementation. If you have a project team and eighteen months, a traditional ERP is a reasonable choice. Most businesses have neither.
No. The write tools are absent from the agent's process rather than denied to it, and every business change needs a recorded human approval.
No, and it never will. It prepares and evidences the return; a person files it. That boundary is deliberate.
Each workspace is isolated at the database level, and tenancy is resolved by the server on every request rather than asserted by the client.
Workspaces are invitation-based: an owner invites their people by email. If you do not have a workspace yet, tell us about your business and we will set one up.
It depends on the size of the business and which of the eleven you switch on. Ask us and we will give you a straight number.
We will come back with what the business actually looks like — customers, best sellers, margin and whatever is hiding in the numbers. You will know inside a day whether this is worth your time.