HX-SDP: Structural Data Platform¶
Release binding: v0.1.0-enterprise-ready · AMI ami-010806d4d3445660e · 2026-05-19
GPU-native. One representation. Every access pattern.
HX-SDP is a single-node GPU appliance that ingests data, represents it as a low-rank latent factorization with an int8 scalar-quantized sidecar, and serves key-value, feature, search, vector, streaming, and columnar access patterns from that one representation. There is no reformatting between workloads and no data movement between stores, because there is one store.
What it consolidates¶
Enterprises assembling an AI data plane from third-party components typically run a KV cache (Redis, Memcached), a feature store (Feast, Tecton), a search index (Elasticsearch, OpenSearch), a vector database (Pinecone, Milvus, Qdrant, Weaviate), an ETL layer to move data between them, an event stream (Kafka) with its retention tier, an API gateway (Kong), a multi-node cluster to hold a billion-vector index, a GPU memory-tiering layer, per-service observability, and a training-data export pipeline. HX-SDP serves those twelve workloads from one signed appliance on one GPU. (Architectural claim; module-level mapping in the architecture reference.)
The archived Google Cloud reference describes an architecture alongside Vertex AI and Google Cloud managed data services. Any cloud or air-gapped deployment must be agreed within the Private Appliance or scoped pilot arrangement. HX-SDP is not a replacement for any Google managed service.
Availability and delivery¶
HX-SDP is currently available through a Private Appliance or scoped pilot. Contact HolonomiX to agree the workload, infrastructure, release bundle, validation criteria, and commercial terms.
The Private Appliance documentation covers VMware OVA, KVM/Proxmox/OpenStack QCOW2, and an air-gapped OCI Runtime Kit for Kubernetes. Use the package and target validated for your agreed release.
The AWS image reference and Google Cloud image reference retain dated technical material. They are not current marketplace purchase or subscription routes.
What you get on first boot¶
- GPU-accelerated retrieval engine (two-tier: latent scan plus SQ8 rescore)
- Multi-tenant gateway with authentication, namespace ACL, rate limiting, usage metering, and audit logging on TLS 8443, the only externally reachable service
- Console UI at
https://<host>:8443/console - Prometheus metrics
- Per-instance ML-DSA-65 (FIPS 204) signing identity and an EXACT retrieval path that returns a signed, independently verifiable receipt
- Signed manifest verified at every boot, fail-closed on tamper
- Offline-verifiable license. No call-home. No automatic data egress