HolonomiX Documentation¶
GPU-native structural data platform. One representation. Every access pattern. No reformatting.
What is HX-SDP?¶
HX-SDP stores data in QTT-compressed tensor format on GPU VRAM. Instead of maintaining separate copies for caching, vector search, feature serving, and full-text search, HX-SDP serves all access patterns from a single structural representation.
| Metric | Value |
|---|---|
| Entries validated | 1B |
| Recall | 100% |
| P50 latency | 38.5 ms |
| Soak test | 24 hours |
| Precision | fp64 / fp32 / fp16 |
Quick links¶
- Quickstart — Store and query in 5 lines
- API Reference — All endpoints with request/response schemas
- SDKs — Python, TypeScript, Go
- Billing — CU costs & tier comparison
- Self-Hosting — Docker Compose deployment
Architecture¶
Client → HX-Gate (auth, rate limit, billing) → HX-Engine (QTT runtime)
↕ ↕
Redis GPU VRAM
(state) (TT-cores)
HX-Gate (port 8080) handles multi-tenant authentication, namespace isolation, rate limiting, CU metering, and Stripe billing. It proxies data requests to one or more HX-Engine instances (port 8000) that run the QTT compression/query runtime on GPU.