Private AI pilot

A private AI pilot for one department

A fixed-price deployment, in four to six weeks, on hardware you own and control.

Book a 30-minute call Vendor due diligence information

Data stays inside your boundary Your documents feed a private AI running on your hardware, used by your users, all inside your boundary. A path to external AI services is blocked. YOUR BOUNDARY Yourdocuments Private AIon your hardware Yourusers Access controls decide who sees what Works offline in air-gapped sites External AI services

What the pilot delivers

On your hardware

AI running inside your own network boundary.

Answers from your own documents

Staff ask a question in plain language and get an answer drawn from the documents you have loaded, with the source shown.

Access controls

People see only what they are already permitted to see.

Handover so your team can operate and review it. Operating guide, admin walkthrough and a 90-minute training session.

Illustrative interface. Document names and answers are examples.

An example of use

A compliance team loads its internal policies. A staff member asks which policy covers sharing customer data with a third party. The system answers in plain language and shows the policy and section it drew from, so the person can check the source.

Only people who already have access to that policy can see it in an answer.

Who can use it

The pilot covers one department, such as compliance, risk, legal or operations. Wider rollout is a separate decision once you have seen it working.

How it runs

Phases and timings are illustrative.

  1. Scope (week 0)
    We agree the department, document set and success criteria in writing.
  2. Environment and ingestion (weeks 1 to 2)
    We install on your hardware and load the agreed documents.
  3. Configuration and testing (weeks 3 to 4)
    We set up access controls and your users test it against real questions.
  4. Handover (weeks 4 to 6)
    Documentation, a walkthrough and a review of what to do next.

For vendor-risk and procurement teams

Due diligence information

The questions a vendor-risk team usually asks, answered directly. A fuller written response is available on request.

What data is processed
Only the documents you choose to load, and the questions your users ask.
Models and licences
Open-weight language models, selected with you for the task. Model names, sources and licence terms are provided in writing before deployment.
Remote access by Maandoh
None by default. A support session happens only when you start it, on your terms.
Logging
Query and access logs are kept on your hardware and belong to you.
Support and incidents
Support by email during Singapore business hours. Response times are agreed in the statement of work.
Exit
You keep the system, your data and the documentation.
Continuity
Maandoh is a one-person firm. See how we handle that.

Comparison

How this differs from AI hosted by a cloud provider

Cloud-hosted AI, including in your own cloud tenant, is a reasonable choice for many organisations. This is where the pilot is different.

QuestionMaandoh private pilotCloud-hosted AI
Where processing runsOn hardware you own, on your premisesOn a cloud provider's infrastructure, in a region and account you configure
Network connectionCan run with no internet connectionNeeds a connection to the cloud service
Who operates the platformYour team, on your hardwareThe cloud provider, under its terms

If your policy allows cloud services and you want to scale quickly, a cloud service may suit you better. The pilot is for cases where policy or risk appetite requires processing to stay on your own premises.

What to expect

What we need from you

  • One department and a named sponsor.
  • Access to the target hardware and an IT contact.
  • A defined set of documents to work with.

Air-gapped sites

The pilot can be installed and operated without an internet connection. Updates are delivered on media you control. No outbound connection is required after installation.

Hardware and price

Hardware
One server with a single data-centre GPU, 128 GB RAM and 2 TB NVMe storage.
Price
Fixed, agreed in writing before work starts. S$38,000

Talk to us

No obligation. We will ask about your environment and tell you plainly whether we can help.

Book a 30-minute call