01
Inference API
OpenAI-compatible endpoints for chat, embeddings and completions with predictable request semantics.
Solution · Intelligent Platform
An AI platform layer your teams can actually ship on
Model serving, RAG Studio, vector search and observability behind one sovereign control plane.
REF · SC-2026-PLAT
Surface · MaaS / PaaS Inference · OpenAI-compatible Retrieval · RAG Studio Telemetry · In-country
Solution · Intelligent Platform
One control plane, fewer integration seams
Teams rarely fail at the model. They fail at the six systems around it. The intelligent platform consolidates serving, retrieval, evaluation, identity and telemetry so a pilot can become a product without a second architecture.
The platform removes the undifferentiated plumbing between a model and a production feature: serving, retrieval, evaluation, access control and telemetry — all inside the same jurisdiction.
- OpenAI-compatible inference so existing SDKs keep working.
- RAG Studio for ingestion, chunking, embedding and retrieval.
- Vector search with sovereign-managed encryption keys.
- Metrics, traces and logs that never leave Malaysian infrastructure.
Capabilities
Platform primitives
The shared services that let product teams ship AI without rebuilding the same plumbing.
02
Model catalog
Choose a supported open model or register your own, with versioning and rollback at the endpoint.
03
RAG Studio
Connect document sources, define chunking and embedding policy, and evaluate retrieval quality.
04
Vector database
Low-latency similarity search with namespace isolation and encryption keys held in-country.
05
Evaluation
Run regression suites against prompts, models and retrieval changes before they reach users.
06
Observability
Metrics, traces and logs for latency, cost and quality, exported to the tools your SRE team already uses.
Process
How a team adopts the platform
A product-oriented sequence that keeps evaluation and governance in the loop from the first sprint.
- 01
Connect
Link identity, document sources and the first environment through the platform control plane.
- 02
Build
Stand up a retrieval pipeline and an inference endpoint against a representative corpus.
- 03
Evaluate
Define quality, latency and cost thresholds, then run the evaluation suite against each change.
- 04
Deploy
Promote the endpoint and retrieval configuration to production with access policy and telemetry attached.
- 05
Observe
Review quality, latency and cost on a recurring cadence and feed regressions back into evaluation.
Use cases
What teams build on it
Product patterns that reuse the same platform primitives instead of a new stack each time.
Internal copilots
Role-aware assistants grounded in controlled knowledge and connected to your identity provider.
Bahasa Melayu support
Customer-facing assistants with BM-native retrieval and escalation to a human agent.
Document intelligence
Extraction and summarisation pipelines with citations back to the source document.
Domain RAG
Vertical retrieval over legal, financial or clinical corpora with evaluation gates.
Controls
Sovereignty controls
Governance that sits with the platform, not in a separate spreadsheet.
In-country processing
Inference, embeddings, vector storage and telemetry are designed to stay in Malaysia.
Key custody
Encryption keys for vector and object storage are managed inside the sovereign environment.
Immutable audit logs
Model, retrieval and administrative events are recorded and exportable for review.
PDPA 2010 alignment
Retention and deletion controls map to Akta 709 obligations for personal data.
FAQ
Questions platform teams ask
Does the inference API work with existing OpenAI SDKs?
Yes. The endpoint follows OpenAI-compatible request and response shapes for chat, embeddings and completions, so most existing SDK integrations work with a base-URL change.
Can we bring our own model?
Yes. Register your weights or select a supported open model. The platform handles versioning, endpoint rollout and rollback, subject to a compatibility and licensing review.
Where does RAG data get stored?
Source documents, chunks, embeddings and the vector index remain in the sovereign environment you designate. Access is governed by the same identity and audit controls as inference.
Can we export telemetry to our own tools?
Yes. Metrics and traces can be exported to your observability stack through standard protocols, with the data path agreed during onboarding.
Solution · Intelligent Platform
Ship on one platform
Bring a use case and a corpus. We will stand up a retrieval and inference path you can evaluate in a bounded pilot.
Quotes are itemised by configuration, term and support. No fixed list price is published.