One governed architecture. Multiple best-fit technologies.

Used in LEGENYX SI engineering, agent workflows, research, and application integrations where appropriate.

Used as part of LEGENYX model-infrastructure and GPU compute workflows for local/cloud inference and serverless execution.

Used for source control, pull-request review, CI/CD coordination, release evidence, and engineering collaboration.

Used in selected LEGENYX and client systems for PostgreSQL data, authentication, storage, edge functions, and realtime capabilities.

Used in LEGENYX enterprise architecture and cloud infrastructure work where Microsoft cloud services fit the system requirements.

Used for selected web application builds, deployments, previews, and production delivery workflows.

Used in selected LEGENYX application builds and rapid product-delivery workflows with LEGENYX architecture, security, QA, and governance layered around the implementation.

Used in selected LEGENYX prototyping and application-development workflows when it is an appropriate delivery surface.

Used in selected LEGENYX engineering, coding, review, research, and agent workflows as one model provider within a model-neutral operating approach.

Used or evaluated for selected LEGENYX web, edge, security, networking, and application-delivery architectures.

A local-first runtime option for running supported models on your own hardware. Deployment choices depend on hardware capacity, model licenses, and project security requirements.

GPU server hardware and inference software can support demanding model workloads. Server sizing, NIM licensing, and Triton deployment are selected around throughput, latency, and operational requirements.

Llama models offer an option for customizable, self-hosted AI workflows. Each model is reviewed for license terms, workload fit, hardware needs, and quality before deployment.

Qwen models can contribute to coding, multilingual, and multimodal workflows. Hosted or self-managed deployment depends on the selected model's license, capabilities, and evaluation results.

DeepSeek offers model options for reasoning and coding tasks. Provider APIs and eligible self-hosted weights are assessed separately for data handling, license terms, reliability, and task quality.

A resource for discovering models and datasets and exploring managed inference. Model provenance, licenses, access controls, and deployment suitability remain part of the engineering review.

A unified API option for accessing models across providers. Routing, fallback behavior, provider data policies, and budgets need explicit configuration for each workload.

A serving engine option for supported self-hosted models. GPU capacity, batching, concurrency, and model compatibility guide whether it fits a production workload.

A desktop option for trying compatible models and exposing a local inference API. Useful for development and evaluation, with deployment and licensing reviewed before wider use.

Gemini models and Google deployment options can support multimodal applications and agent workflows. API choice, data policies, region availability, and quality are evaluated per project.

Mistral provides hosted models and selected open-weight options for language and coding applications. Model-specific licenses, hosting requirements, and task performance determine the fit.

A hosted inference option for supported models when responsiveness matters. Model availability, request limits, data handling, and measured latency guide workload selection.