Join the pilot waitlist

AI customer support that runs on your own server.

No cloud AI provider, no per token bill. Built to run on your own server, for Singapore online stores. Pilot waitlist now open.

Watch it answer, live

Unedited. The chat widget on a real Shopify store, answering three questions a customer might actually ask.

The same questions, all day

  • Where is my order?
  • How long does delivery take?
  • Can I return this if it does not fit?
  • Does this come in a bigger size?
  • Can I change my delivery address?

Your team already knows the answers. Most of them sit in your policies, product pages and FAQs already. Someone still has to type them out again, ticket after ticket.

Two worries, answered by how it is built

A decision you can skip
  • Customer questions, order details and internal policies do not have to pass through a third party's AI service.
  • Hapio has no cloud AI provider, no API key and no HAPIO server in the data path.
  • That is how it is built, not a promise about how we behave.
  • Your customers' names, addresses and orders are personal data, and the PDPA is why you think twice about where they go.

A meter you never have to watch
  • Most hosted AI is billed per token, so the bill grows with every customer message.
  • Hapio has no AI vendor, so there is no per token meter.
  • What it costs you to run is your own hardware.
  • It is not a bill that rises when a sale weekend brings a rush of questions.

How it works

  1. It runs on your server

    You install it with Docker on a machine you control that has an NVIDIA GPU. There is no HAPIO server in the middle and no telemetry.

  2. It answers from your documents

    Give it your policies, product information and FAQs. Each answer shows the passages it came from, so anyone can check it. It works from documents, not from your order system, so it does not look up individual orders.

  3. It goes on your site with one snippet

    Paste one line of code into your website and the chat widget appears.

Both models are open weights and both run on your hardware. qwen3:8b writes the answers and bge-m3 searches your documents. It retrieves passages for each question. It does not retrain on your documents.

What you need

Hapio needs a real server with a GPU, not a laptop. If that is not something you run today, this pilot is probably too early for you, and we would rather tell you now than later.

  • GPU 24 GB VRAM E.g. RTX 3090, RTX 4090, NVIDIA L4
  • CPU 16 cores
  • RAM 32 GB
  • Disk About 60 GB free
0.66s Median response time across a 12 hour continuous test.
16 to 24 Simultaneous chats with flat latency, tested to 128.
Zero Errors across the entire 12 hour test.

If the machine runs other programs too

  • Add what those programs use on top of the figures above, for both system RAM and disk.
  • Anything else that uses the GPU, such as a desktop session, video work or another AI model, draws from the same VRAM budget, leaving room for fewer chats at once.
  • Where possible, give Hapio its own machine, or at least its own GPU.

Join the pilot waitlist

Hapio is not for sale yet, and the pilot is small. Tell us about your store and your setup. Applicants whose setup fits the pilot will be asked for a short call.

Your given name.
Your family name.
A number we can call to arrange the short call.
We will write to you here. An address at your store's own domain works best.
The registered business name, if it differs from your store name.
Your Singapore Unique Entity Number: eight or nine digits followed by a letter.
Everyone who works at the company, in total.
The name customers see.
The address of your online store, starting with https://
Hapio needs an NVIDIA GPU. Pick the closest match, including "No server yet" if that is the truth.
When you would want to try it. Just exploring is a fine answer.

Terms of Use  |  Privacy Policy  |  Acceptable Use Policy  |  Software and Distribution Terms

Q&A

Is it for sale?

Not yet. We are running a small pilot first, for Singapore online stores only. The form takes no payment. Applicants whose setup fits the pilot will be asked for a short call.

Does my data leave my server?

The models that read your documents and write the answers run on your own hardware. There is no cloud AI provider, no API key and no HAPIO server in the data path, and Hapio sends no telemetry. Your server still has to be reachable by your customers' browsers, because that is where the chat widget talks to.

What hardware?

A server you control with an NVIDIA GPU (24 GB of VRAM recommended), at least 16 GB of system RAM, and Docker. The answering model is qwen3:8b and the search model is bge-m3. The full breakdown is under What you need, and we will confirm exact requirements with pilot applicants whose setup fits.

Does it make mistakes?

Yes, it can. It answers from your documents and shows its sources, so a person can check them. Where your documents are thin or unclear, it can still get an answer wrong. Have a person review its answers, especially on refunds and anything involving money.

Who is behind it?

CR8TEK (UEN 53531114A) is a sole proprietorship registered in Singapore at 60 Paya Lebar Road, #06-28 Paya Lebar Square, Singapore 409051. CR8TEK operates HAPIO. We are running this pilot to find out whether Singapore online stores want it before we build more.