API Reference¶
HX-SDP exposes a RESTful HTTP API. All data-plane traffic flows through the HX-Gate reverse proxy, which handles authentication, namespace isolation, rate limiting, metering, and audit logging before forwarding to the HX-Engine backend.
Base URL¶
| Environment | URL |
|---|---|
| Production | https://gate.holonomx.com |
| Self-hosted | http://<host>:8080 |
Authentication¶
Every data-plane request must include a tenant API key:
or
Admin endpoints use the engine service key instead. See Authentication for details.
Request / response format¶
- Content-Type:
application/jsonfor all JSON endpoints. - Binary payloads: Base64-encoded in the
data_b64field. - All responses are JSON with appropriate HTTP status codes.
Common headers¶
| Header | Direction | Description |
|---|---|---|
X-Api-Key |
Request | Tenant API key |
X-Request-Id |
Both | Correlation ID (gate injects if absent) |
X-HX-Tenant |
Request (injected) | Tenant ID injected by gate into engine requests |
X-HX-CUs |
Response | Compute Units consumed by the operation |
Verb model¶
HX-SDP uses five core verbs:
| Verb | Endpoint prefix | Description |
|---|---|---|
| PUT | POST /v1/put |
Ingest — compress and store a tensor or blob |
| GET | GET /v1/get/{ns}/{key} |
Retrieve metadata and compression fingerprint |
| QUERY | POST /v1/query/* |
Similarity search — pairwise, top-K, or external vector |
| SEARCH | POST /v1/search |
Metadata-filtered key listing |
| SERVE | POST /v1/serve |
Reconstruct dense array or load TT-cores to GPU VRAM |
Additional administrative verbs are documented in the Admin section.
Error responses¶
All errors return a JSON body:
| Status | Meaning |
|---|---|
400 |
Malformed request (missing key, bad JSON, invalid namespace) |
401 |
Missing or invalid API key |
403 |
Namespace not authorized for this tenant |
404 |
Key not found in the specified namespace |
409 |
Version conflict (optimistic concurrency) |
429 |
Rate limit exceeded — retry after window resets |
502 |
Engine backend unreachable |
503 |
Monthly CU quota exhausted |
CU metering¶
Every operation consumes Compute Units. The X-HX-CUs response header reports the cost. See Billing & Usage for the full cost table and tier definitions.
Pagination¶
List endpoints return up to 1,000 results by default. Use ?limit=N&offset=M query parameters for pagination.
Next steps¶
- Storage API —
PUT,GET,SERVE,DELETE,LIST - Query API — pairwise similarity, top-K, vector search, metadata search
- Simulation API — physics simulation dispatch
- Admin API — tenant management, usage, registry reload