by

Private cloud

Start in the cloud, move in later

Your own Core on a dedicated cloud server we run for you. Running in days, with no machine to keep, and ready to move to your office.

Request a demoSee pricing
A film of Core, on demo data. The morning Brief says the book was up 8 bps overnight, and a citation opens the analyst note behind it. Asked where our NVIDIA preview sits against consensus, Core searches S&P Global Visible Alpha and answers with a chart, then lists its six sources: licensed data, filings and the firm’s own notes. One source opens in the Library with the cited sentence highlighted. Then a rumor that Microsoft cut its GB300 order is weighed against the firm’s evidence: likely true, 84%.

The cloud or your office, row by row.

  • Where it runs

    Private cloud
    A dedicated cloud server, one per firm, in a region we agree
    On-premises
    A machine on your premises, on your network

  • Who owns the hardware

    Private cloud
    The cloud provider. We run the server
    On-premises
    You

  • Who has admin access

    Private cloud
    Our engineers, through the cloud account
    On-premises
    Your IT team, and our engineers when you let them in

  • The model

    Private cloud
    Claude, GPT or Gemini, through their APIs
    On-premises
    An open model on your GPU. A cloud model if you switch it on

  • The search index

    Private cloud
    Built with a cloud embedding model, so document text reaches the provider
    On-premises
    Built on the machine

  • How people reach it

    Private cloud
    Any browser, at your Core’s web address, behind sign-in
    On-premises
    Your office network, or a relay you control

  • Upgrades

    Private cloud
    We deploy each release to your server
    On-premises
    From a signed release or a disk, with rollback in one command

  • Backups

    Private cloud
    Scheduled backups we set up, kept in your region
    On-premises
    Scheduled backups we set up, to a disk or share you choose

  • Time to start

    Private cloud
    Days
    On-premises
    Once the machine is in place

  • What you pay for

    Private cloud
    A server each month, and model tokens
    On-premises
    The machine once. No charge per question on an open model

What AWS and Microsoft add, when we set it up.

No operator access
On AWS Nitro servers, AWS states there is no way for its staff to log in to the host or read memory, a design NCC Group reviewed in 2023. Azure confidential VMs make the same claim for Microsoft
Keys your firm holds
Disks and backups encrypted under keys you control: AWS KMS or keys kept outside AWS, or Azure Key Vault and Managed HSM. Revoke the key and the data cannot be read
No open ports
Admin access through AWS Session Manager or Azure Bastion, each session logged, nothing open to the internet
The model in your cloud
Claude on Amazon Bedrock in Seoul or Singapore keeps prompts in that region and away from the model provider. Azure OpenAI keeps prompts from OpenAI
In-country regions
Hong Kong, Singapore and Seoul, on both clouds
A longer audit trail
The cloud’s own audit log, kept past its 90-day default

These are AWS’s and Microsoft’s published claims, and they apply once configured. We scope the ones your firm needs in your proposal. None of them cover a model called through an outside API.

Why firms start here.

No machine to buy
Nothing to order, house or power
Running in days
A server and your Core, set up and reading your sources within the week
Frontier models
Claude, GPT and Gemini for every answer
From anywhere
Your team signs in from any browser, in the office or out
We run it
Releases, certificates, restarts and the server itself
Try before hardware
Learn what your desk needs before you buy a machine

What you give up.

Your data leaves the office
It lives on a server we run in the cloud, not on your premises
Text reaches the model provider
Questions, passages and document text for the search index go to the provider of the model
Outside your network
It is protected by sign-in, not by your office walls
More rules apply
Cloud storage of records falls under the SFC’s external storage rules, FSS reporting and MAS outsourcing guidelines
Region and model limits
Claude on Bedrock is not yet in the Hong Kong region. An open model needs a GPU server in the cloud
Costs grow with use
Model tokens are billed for every question

Move in when you are ready.

  1. 01

    The machine arrives

    A DGX Spark we supply, or your own server.

  2. 02

    We copy your Core

    The library, the map, the answer log and every setting.

  3. 03

    Re-index on the machine

    Search is rebuilt with the model on the machine. Allow a day for a large library.

  4. 04

    Switch over

    Your team signs in at the same address. We close the cloud server.

Questions.

Is it as private as on-premises?

No. On-premises keeps everything in your office. In a private cloud your documents live on a server we run for you, and text goes to the model provider. Choose by what may leave your firm.

Which cloud and region?

AWS today, in Seoul. Hong Kong and Singapore regions, or Microsoft Azure, on request.

Is any other firm on our server?

No. One server per firm, with its own library, answer log and sign-in.

Whose model key is used?

Yours, if your firm has an agreement with Anthropic, OpenAI or Google. Otherwise ours, scoped in your proposal.

Can we move it to our own machine later?

Yes. We copy your Core, its library and its answer log to a machine in your office, and rebuild search there.

Explore more of Core.

Start without a machine.

Tell us your firm and your sources. We reply within one business day.

Request a demo

© 2026 Treasurer Global Pte. Ltd. All rights reserved.

Member of the NVIDIA Inception Program