Installation Requirements
These requirements must be satisfied by the infrastructure or private cloud in which the Hiya Service is to be deployed.
Container runtime support
The infrastructure must be capable of running containers on any of the supported container runtimes and supported CPU architectures.
This is ideally achieved with a container orchestration platform such as:
- Kubernetes
- OpenShift
- Docker Swarm
All major cloud providers offer services that satisfy this requirement:
| Cloud Provider | Service |
|---|---|
| Amazon Web Services | Elastic Container Service (ECS) |
| Microsoft Azure | Azure Kubernetes Service (AKS) |
| Google Cloud Platform | Google Kubernetes Engine (GKE) |
The Hiya Service can be deployed in the same cloud or cluster as the client application that will consume it.
Hardware requirements
Each engine instance requires the following resources:
| Resource | Minimum | Recommended | Notes |
|---|---|---|---|
| RAM | 5 GiB | 8 GiB | ~5 GiB for Triton + models in memory; 8 GiB provides headroom for concurrent requests |
| CPU | 4 cores | 8 cores | Peaks at ~2.5 cores during inference, idles near 0. 8 cores recommended for reasonable latency |
| Disk | 4.5 GiB | 4.5 GiB | For the Docker image (ML models + Triton + engine binary) |
| tmpfs | 8 GiB | 8 GiB | RAM-backed filesystem for models during startup |
The tmpfs mount shares physical RAM with the engine process. The recommended 8 GiB of total RAM accommodates both the tmpfs allocation and engine overhead.
Network connectivity
Connectivity to api.hiya.com:443 is required for licensing and billing based on pay-per-use. No audio or sensitive data is transferred.
- Communication is performed using HTTPS over port 443.
- If required, a point-to-point VPN using IPSECv2 can be established to further secure the communication.
See the Network Connectivity Matrix in the Reference for a complete list of required connections.