What FactSmith itself costs
This page prices the machine, not the licence. FactSmith is sold per deployment, to one company at a time, with no per-user charge. We give you a real number on a call rather than post a figure that changes next month. Ask for it.
The work is answering questions
Storage is not the bill. The machine earns its keep when someone asks.
FactSmith is two programs on one host: the application you open in a browser, and a small database of saved answers. Your warehouse stays where it already is. When someone asks, that host transcribes speech if they spoke, turns the question into a query, and runs it. The middle step needs a model. Everything else is modest.
Saved dashboards and facts are small. Disk is not what you are sizing for.
Three ways to host it, same rule
Cloud, small computer
Rent an ordinary machine. The application runs there. A model you have a key for answers from outside. The monthly number does not change when the tenth person starts asking.
Cloud, with an open model
Rent a machine with a modest graphics processor and run an open model in your own cloud account. Questions do not go to an AI company. Still no per-person charge.
A box you own
Buy a machine, load an open model, keep every question inside the building. After you own it, another user is not a line on a bill.
None of these is a per-person licence. That is the point of a single-company deployment.
What the machine actually needs
Sized for processing, not for storing files.
The evaluation copy already runs on an ordinary Windows PC with 16 GB of memory, including the voice transcription. That is the floor. It gets slow if the same machine is also running everything else you do.
An open model that answers well enough for business questions fits in a few extra gigabytes. A modest graphics processor (the kind sold for a gaming PC) makes those answers snappy. You do not need a datacentre card. Questions arrive one at a time, and the processor sits idle between them.
Several people can use it at once. They share one machine. For a business that is not asking hundreds of questions in the same second, that is the right size.
If the model runs somewhere else, on a key you supply, the host can stay a small cloud computer. That is a different bill: per question, still not per person, and the question leaves your network. The product shows that before you choose.
Cloud, in round numbers
Priced in US dollars, because that is how cloud computers are sold, with a rough rand band alongside for budgeting. Figures as of August 2026. They move.
| What you rent | What it does | Order of cost (USD) | Order of cost (ZAR, roughly) |
|---|---|---|---|
| Small computer, 16 GB | The application. Model answered elsewhere, or slowly on this machine. | About a hundred dollars a month | Roughly R1,800 a month |
| The same, with a modest graphics processor | Open model in your own cloud account. Questions do not go to an AI company. | A few hundred dollars a month | Roughly R3,500-R9,000 a month |
These are public list prices for ordinary cloud computers, not a FactSmith quote. Rand figures are a rough conversion at an August 2026 exchange rate, not a local quote: your region, discount, and the rand/dollar rate on the day will all differ. Storage is omitted on purpose.
Hardware, in round numbers
Bought once. The running cost after that is electricity.
| What you buy | What it does | Order of cost |
|---|---|---|
| The PC you already have | Evaluation, and light use on this machine. | Nothing extra |
| A small workstation or mini PC, 32 GB | Comfortable open models without a graphics card, still offline. | Low thousands, once |
| The same with a modest graphics card | Fast open models. Fully offline. | Low to mid thousands, once |
A few years of renting a cloud machine with a graphics processor is the price of owning one. After that the cloud meter is still running and the box is paid for.
Why offline is the better default
The product can run with nothing leaving your network. That is a choice, not a slogan.
- Questions, spoken audio, and the names of your products stay in the building.
- No internet is required after the copy has reached you.
- An open model on your hardware does not have a vendor who can change the price of a question, or withdraw the model.
- The tenth person to ask is not a new licence. The hundredth is not either.
- When the line is down, the answers are not.
Cloud is faster to start, and a hosted model is often faster than a small office PC. You see that trade in the product. Offline is the path where a question’s processing cost falls to electricity, and where privacy does not depend on a vendor’s terms.
Where to go next
What standing one up involves: Deployment. What is enforced underneath: Trust and security. Common questions: FAQ.