DXC builds multi-model AI operations platform on Amazon Bedrock
DXC Technology has built a multi-model foundation for its DXC OASIS AI operations platform on Amazon Bedrock, combining access to several model families with infrastructure-level security and governance. The design is intended for mission-critical environments operated by banks, hospitals and government agencies.
- DXC OASIS uses Amazon Bedrock to access Anthropic Claude, Amazon Nova, Meta Llama and open-weight models.
- Amazon Bedrock Guardrails apply filtering, PII redaction and denied-topic policies across models.
- AWS PrivateLink keeps inference traffic within the customer’s virtual private cloud.
- DXC reached general availability for OASIS in March 2026.
- AWS says resolution time fell by 95% on average across 56 measured use cases.
A governed gateway to multiple models
According to AWS, DXC OASIS reaches Anthropic Claude, Amazon Nova, Meta Llama and open-weight model families through Amazon Bedrock model inference. A single managed interface allows DXC to change model routing without rebuilding the system if a provider has an outage or deprecates a model.
Amazon Bedrock Guardrails apply content filtering, personally identifiable information redaction and denied-topic policies across the models. AWS PrivateLink keeps inference traffic within the customer’s virtual private cloud, while serverless scaling removes the need for DXC to manage GPU infrastructure for the model layer.
The platform also keeps customer data within the AWS data boundary. AWS says prompts and completions are not shared with third-party model providers and are not used to train the underlying foundation models. Model availability varies by AWS Region.
Separating the substrate from the AI harness
DXC divides OASIS into two layers. Amazon Bedrock provides the governed model substrate, while DXC supplies the harness that controls how work is carried out. That upper layer includes orchestration across tools and systems, operational context based on 60 years of runbooks and escalation logic, policy enforcement, human-oversight gates and auditability for agent actions.
The system runs its agent runtime on Amazon Elastic Kubernetes Service in a private subnet. A supervisor agent delegates work to capability agents that can use a knowledge base, AWS Secrets Manager, Amazon EventBridge and Amazon Simple Storage Service. Calls to models leave the subnet through a VPC interface endpoint and AWS PrivateLink rather than traversing the public internet.
AWS Identity and Access Management, AWS Key Management Service, TLS 1.2, Amazon CloudWatch, VPC Flow Logs, AWS CloudTrail and OpenTelemetry provide security and observability across the architecture.
Routing work according to capability and cost
DXC began building OASIS as a prototype in early 2025 and reached general availability in March 2026. The company is scaling toward 130 customer environments on Amazon EKS, using per-tenant namespaces and GitOps rollout.
Model selection is treated as a workload decision. DXC can route tasks according to cost, latency and capability, rather than being constrained by a single model provider. Intelligent prompt routing sends simpler prompts to lower-cost models and complex reasoning tasks to more capable ones, while prompt caching is intended to reduce latency and cost. Custom Model Import extends the same governance and tooling to models supplied by DXC or its clients.
Before a model enters production routing, the OASIS harness evaluates it against task suites. AWS says that, since general availability, DXC OASIS has reduced resolution time by 95% on average across 56 measured use cases.
For users in regulated industries, the architecture provides a way to combine model choice with private networking, regional inference controls and policy checks. AWS describes the substrate as the infrastructure foundation, while DXC remains responsible for the orchestration, assurance and human approvals that determine how agents operate.
