01
Private connectivity
Connect over a dedicated circuit or IPsec VPN. The model endpoint is not exposed to the public internet.
Solution · Enterprise AI
Private AI for the enterprise, without the public-cloud detour
Private model deployment, retrieval and inference behind your network boundary — with sovereignty controls your auditor can read.
REF · SC-2026-ENTAI
Deployment · Private Connectivity · Dedicated / VPN Identity · SSO / SCIM Data plane · In-country
Solution · Enterprise AI
AI behind your boundary, not beside it
Enterprise AI fails procurement when the model is useful but the data path is unacceptable. We deploy the model, retrieval layer and gateway as one private system, then connect it to your identity and network on your terms.
Bring your own model or use a supported open model. Either way, prompts, documents, embeddings and outputs stay inside the environment you control.
- Private connectivity over a dedicated circuit or IPsec VPN.
- Tenant and subnet isolation with no shared model-serving surface.
- SSO, SCIM and role-based access tied to your existing directory.
- Immutable audit logs for prompts, admin actions and data access.
Capabilities
The private AI stack
Model serving, retrieval, identity and governance deployed as one controlled environment.
02
Network isolation
Private subnets, security groups and tenant boundaries separate your workloads from every other environment.
03
Private model serving
Serve a supported open model or bring your own weights on an OpenAI-compatible endpoint.
04
RAG and document AI
Ingest, chunk, embed and retrieve your own corpus with access controls carried through to the answer.
05
Identity and SSO
SAML 2.0, OIDC and SCIM provisioning integrate with Okta, Entra ID or your sovereign identity provider.
06
Governance and audit
Prompt, retrieval and administrative events are logged immutably and exportable to your SIEM.
Process
From pilot to private production
A staged path that proves value and control before a full production commitment.
- 01
Discover
Define the use case, data classification, success metric and the regulator whose expectations must be met.
- 02
Design
Agree the model, retrieval design, connectivity path, identity integration and isolation boundary.
- 03
Pilot
Run a bounded pilot against a representative corpus with named evaluation criteria and an access review.
- 04
Production
Harden, scale and connect the pilot to production identity, monitoring and incident processes.
- 05
Operate
Review quality, cost and access on a scheduled cadence, with a documented change path for models and prompts.
Use cases
Private AI patterns
Workloads where the value is high but the data path has to survive an audit.
Internal knowledge assistant
Answer policy, HR and operations questions from controlled internal documents with citations.
Document processing
Extract, classify and summarise contracts, claims and records without sending them offshore.
Contact-centre support
Assist agents with Bahasa Melayu and English responses grounded in approved knowledge.
Engineering copilots
Offer code assistance against private repositories while keeping source code in-country.
Controls
Sovereignty controls
Controls procurement, security and compliance teams can inspect before signing.
In-country data plane
Prompts, documents, embeddings and outputs are designed to remain in Malaysia.
PDPA 2010 alignment
Data-subject access, retention and deletion procedures map to Akta 709 obligations.
Identity-bound access
Every request is tied to an authenticated identity and an authorisation policy.
Audit-ready records
Administrative changes, retrievals and model calls are recorded for internal or regulator review.
FAQ
Questions security teams ask
Is our data used to train the model?
No. Customer documents, prompts and outputs are not used to train shared models. Any fine-tuning on your corpus is performed only on your instruction and remains your property.
Can we bring our own model weights?
Yes. You can deploy supported open models or bring your own weights, subject to a compatibility and licensing review. The serving endpoint remains OpenAI-compatible.
How does the platform connect to our network?
Through a dedicated circuit or an IPsec VPN, with private subnets and security groups on our side. The endpoint is not exposed to the public internet unless you explicitly request it.
What identity providers are supported?
SAML 2.0, OIDC and SCIM 2.0. We have integration patterns for Okta, Microsoft Entra ID and sovereign identity providers, and will validate your IdP during the design phase.
Solution · Enterprise AI
Put the model behind your boundary
Bring the use case and the data classification. We will return a private deployment design with the connectivity and identity path.
Quotes are itemised by configuration, term and support. No fixed list price is published.