HX-SDP — Structural Data Platform
One representation.
Every access pattern.
GPU-native runtime built on the HolonomiX engine. QTT-compressed storage, sub-millisecond retrieval, verified physics simulation, and cryptographic evidence on every result. No reformatting. No copying.
1B
Entries Validated
100%
Recall
38.5ms
P50 Latency
24hr
Soak Test
Platform
Storage
QTT-Compressed Vectors
Embeddings and tensors stored in Quantized Tensor Train format. Cores remain on GPU. No decompression in the query path. No accuracy loss.
Search
Billion-Scale Similarity
Top-K nearest neighbor search operating entirely in compressed domain. Cosine, dot product, L2 metrics. Metadata filters. Sub-millisecond at scale.
Simulation
Verified Physics
Navier-Stokes, Vlasov-Maxwell, Maxwell 3D. Every result validated against conservation laws and trust-certified. Reproducible by construction.
Security
Multi-Tenant Isolation
Namespace-level access control. Per-tenant rate limiting, CU quota enforcement, JSONL audit trail. SHA-256 key hashing. Zero-trust architecture.
Compute
GPU-Native Runtime
Tensor Train cores live on GPU from ingest to retrieval. Triton-fused kernels for contraction, rounding, and search. No CPU round-trips in the hot path.
Integration
TypeScript / Go / Python
Production SDKs with typed models matching the wire protocol exactly. Install, authenticate, ship. Full simulation control from any stack.
Five lines to first query
Install the SDK. Authenticate with your key. Store vectors in compressed TT-core format. Search at GPU speed. No infrastructure to manage.
Full documentation →// npm install @holonomix/sdk import { HolonomiXClient } from "@holonomix/sdk"; const hx = new HolonomiXClient({ baseUrl: "https://gate.holonomx.com", apiKey: "hx_live_...", }); // Store await hx.putCores({ key: "vec-001", cores: cores.map(c => ({ data_b64: c })), }); // Search const { results } = await hx.queryTopK({ query_key: "vec-001", k: 10, });