Question 1

What does Refactr actually do?

Accepted Answer

We run AI models on computers you control, so your team can use AI without your information going to an outside company. You choose the model, we install the software that makes it fast and reliable on your hardware, and we keep it running afterwards. We do not build apps or touch your documents; we make the AI underneath work.

Question 2

Is this only for large enterprises?

Accepted Answer

No. It suits any company where a few dozen people or more would use AI every day, or where the information involved cannot go to a public service. A single server with two graphics cards is enough to start; our published benchmarks run on exactly that.

Question 3

What will our staff use it for?

Accepted Answer

The same things they would use a public AI tool for: asking questions of company documents, summarising and drafting, help with code, and assistants for internal processes. It looks and feels like the tools people already know. The difference is that it runs in your environment.

Question 4

Do we need an AI team or a developer to run it?

Accepted Answer

No. We install it, monitor it, update it and support it. Your IT contact keeps ownership of the hardware and the data and can see everything we do. Nobody on your side has to become a specialist.

Question 5

Is this cheaper than paying per seat or per request?

Accepted Answer

For daily use across a team, yes. Public AI services charge for every seat or every request, so the bill grows with use. A model on your own hardware costs the same whether ten people use it or two hundred, and our software gets more work out of each machine than standard setups do. We measure it on your hardware before you commit.

Question 6

How do you charge?

Accepted Answer

The pilot is free for fourteen days: setup on your GPUs in about a week, then measured against what you pay today. If the numbers hold, you pay a setup fee and a monthly fee for running it, quoted for your setup after a short call. Because the model runs on hardware you own, what you pay does not rise with how much your team uses it.

Question 7

Do we have to buy new hardware?

Accepted Answer

Usually not. We start with what you already have, including workstation and gaming-class graphics cards, from AMD as well as NVIDIA. If more capacity is genuinely needed we tell you exactly what and why, and we can source it for you.

Question 8

What if it does not work out?

Accepted Answer

You find out in the first two weeks, not after a long project. The pilot runs on your hardware with your model and reports speed and cost against what you pay today. If the numbers hold, you roll it out wider. If they do not, you stop there.

Question 9

Does our information ever leave our environment?

Accepted Answer

No. The model runs on your servers, in your building or in your own private cloud account, and every question, document and answer stays there.

Question 10

Can Refactr staff see our documents or conversations?

Accepted Answer

Not unless you let us in. Nothing is sent back to Refactr; there is no usage reporting or telemetry. When you want support, you grant us access for that session and can revoke it afterwards.

Question 11

Is our data used to train anything?

Accepted Answer

No. Nothing your team types or uploads is used to train the model or anyone else's. The model you deploy is fixed until you decide to change it.

Question 12

Where does this leave us on GDPR, HIPAA or similar rules?

Accepted Answer

Better placed than with a public service, because the data never leaves your control, which is the core of what those rules ask for. We do not hold certifications ourselves; the system runs inside your environment and under your existing controls, and we can walk your compliance lead through exactly how.

Question 13

How long until people are using it?

Accepted Answer

Under two weeks from the first call, for a pilot with a small group on your hardware. Wider rollout follows once the pilot has shown the speed and the cost you expected.

Question 14

Which AI model will we use, and can we change it later?

Accepted Answer

Any compatible open model, and yes. Most businesses start with a leading open model that runs well on their hardware; if you already run one you trust, we serve that. Switching models later is a normal change, not a new project. Refactr does not sell models and is not tied to a vendor.

Question 15

Does it work with the tools we already use?

Accepted Answer

Yes. Epsilon exposes one endpoint on your network, and your existing tools point at it instead of at a public service. We set that up as part of the deployment; anything that already talks to an AI service can talk to yours.

Question 16

What happens when something needs updating?

Accepted Answer

We handle it. Software updates, performance tuning as usage grows and model changes you ask for are part of the ongoing service, scheduled with you and tested before they go live.

Question 17

What support do we get after launch?

Accepted Answer

Monitoring, updates and a direct line to the people who set it up. If something is slow or stops, we see it and fix it. Your team keeps control of the hardware and the data throughout.

Make private AI earn its hardware.