What is shaide?¶
shaide is a self-hosted AI platform that serves models at scale on your own Kubernetes clusters. It runs many models side by side, each with multiple replicas, and routes traffic across them.
The problems it solves¶
Assembly. Building enterprise AI infrastructure means selecting and integrating an inference engine, a serving orchestrator, a gateway, a model registry, storage and observability - then repeating it per environment. shaide ships that stack as one installable platform.
Sovereignty. shaide runs entirely inside your perimeter. Nothing phones home, and there is no dependency on a vendor cloud - up to fully air-gapped clusters with no internet access.
What you get¶
| Capability | Detail |
|---|---|
| Distributed inference | vLLM engine, llm-d multi-instance orchestration |
| Multi-model serving | Generative and embedding models concurrently |
| OpenAI-compatible API | Existing SDKs and agents work unchanged |
| One-command install | Self-contained interactive installer |
| Air-gap support | Internal registry mirrors all images and weights |
| Infrastructure as code | Every layer a Pulumi program |
What it is not¶
- Not a hosted service - you run it.
- Not a cluster provisioner - it installs onto an existing cluster.
- Not a model trainer - it serves models, it does not fine-tune them.
- Not an inference engine - it orchestrates vLLM, but does not implement the model runtime itself.