NEXUS AI PLATFORM / DEPLOYMENT
The same platform, wherever you need it to run.
One container image runs on every target; only the orchestration layer differs. Which one you pick is a question about your operations team and your data-residency position, not about what the platform can do.
Pick by operations appetite, not by capability. Nothing on this list is a reduced version of the platform.
- Kubernetes
- Best when: You already run a cluster and want fine-grained control across several services
- Managed container services
- Best when: You are on a major cloud and want the smallest possible operations footprint
- A single host
- Best when: Development, a demo, or a small single-tenant deployment
- Fully on premises
- Best when: Data residency is non-negotiable and nothing may leave the estate
- Hybrid
- Best when: Some components have to stay inside and others are better run in your cloud
| Option | Best when |
|---|---|
| Kubernetes | You already run a cluster and want fine-grained control across several services |
| Managed container services | You are on a major cloud and want the smallest possible operations footprint |
| A single host | Development, a demo, or a small single-tenant deployment |
| Fully on premises | Data residency is non-negotiable and nothing may leave the estate |
| Hybrid | Some components have to stay inside and others are better run in your cloud |
Nexus is model-agnostic, and different groups can run different models at the same time. Profiles are how that gets organised instead of becoming a per-team argument.
Default
General traffic. The balance of answer quality and cost that suits most of what people ask in a day.
Fast
High-volume, latency-sensitive, straightforward work, where an answer that arrives immediately is worth more than the last few points of nuance.
Thoughtful
Complex, high-stakes, multi-step work — and anything the platform cannot take back once it has done it.
Which model sits behind each profile is decided with you, against the models you already pay for or want to adopt. Model availability moves faster than a website does, so the mapping is settled in the technical session rather than published here. Profile behaviour comes from MagikDev's own testing, not from a measurement of your environment, and the right mix is set against your real traffic.
Retrieve only what a question needs
A narrow question does not pull the whole estate into context in order to answer it.
The right model for the task
A simple, high-volume lookup does not pay for reasoning it never uses.
Deterministic where it can be
Work that does not need a model does not go to one.
- Sizing against your real job volumes and user counts
- Network requirements
- Identity integration
- How the AI components run inside your estate
- Backup and recovery
There is no reference figure on this page, and that is deliberate. We would rather work these through against your actual numbers than publish a sizing that will not match your environment and then spend the first meeting explaining why.
A technical session covers your estate, where the platform would run, and which pods come first.
Book a technical session