Portable runtime · Cloud

How UI-authored and cross-language dataset definitions run in hypequery cloud, and the execution capability matrix.

hypequery cloud is not live yet

You can't sign up for or deploy to cloud today. These pages describe the planned service, and details will change before launch. What runs today is the open-source TypeScript and Python SDKs, in your own backend.

The UI can author a portable deployment definition directly. Validate it using the protocol package, then load it with rehydrateProtocolDeploymentContract. TypeScript and Python authors export the same deployment vocabulary. Cloud's portable runtime rebuilds these definitions and uses the TS dataset compiler; Python local execution uses its own native planner.

Providers can use createPortableSemanticRuntime({ queryBuilder: db }) for both runtime.compile(input) and runtime.execute(input). Both resolve the same activation catalog, tenant context, and row budget. These are trusted adapter operations: the deployment data plane or an equivalent provider boundary must authenticate and authorize the request before supplying the input. Compilation does not run preflight guard statements; execution runs them before reading rows. Portable compilation includes activationRevision in its structural description so the diagnostic record identifies the definition generation used.

Use the shipped capability matrix to enable authoring controls for a target:

import { DATASET_EXECUTION_CAPABILITIES } from '@hypequery/datasets';

const cloudFeatures = DATASET_EXECUTION_CAPABILITIES.features
  .filter(feature => feature.cloud)
  .map(feature => feature.feature);

The matrix distinguishes protocol representation, TS/Python local support, TS/Python publication, and repository cloud execution. It covers base and derived measures, segments, window/shift measures, approximate distinct counts, and sub-day grains. It is a conservative feature discovery aid, not per-definition validation or a statement that a hosted service has qualified Python deployments.

Publishing currently emits deployment v2. Its formula grammar is narrower than local planning: a derived measure may reference only base measures, and a literal may appear only as round decimals or a coalesce fallback. buildProtocolDeploymentContract and assertDeployable refuse anything else with an error that names the dataset and measure; see Deployable formulas. Protocol v3 acceptance alone does not enable v3 features in the portable runtime. When adding a portable dataset feature, complete its protocol representation, SDK authoring/export, cloud rehydration and execution, and Python native planning; update the matrix and its TS/Python probes together. Enable UI controls only for a target whose executable support is present.