@0xintuition/atom-runtime

Shared Promise-based composition for classification and enrichment. Requires Node.js 20 or newer.

import { createProcessingRuntime } from '@0xintuition/atom-runtime'

const runtime = createProcessingRuntime({
  env: {},
  cache: { provider: 'memory' },
})
const classification = runtime.classification.engine()
const enrichment = runtime.enrichment.engine('default')
// Use the engines' classify/enrich methods with your application's requests.
runtime.dispose()

createProcessingRuntime({ config, runtime, overrides }) also accepts explicit clock, logger, cache fetcher and private plugin overrides. Engines and caches are shared within a runtime. Enrichment instances are memoized by preset and identity capability. Dispose prevents new engine lookups; applications remain responsible for cancelling requests already in flight.

Public subpaths: presets, fingerprint, consumers, config, adapters/upstash, and package.json. The root exports configuration schemas, pickProcessingEnv, preset factories, defaultScopes, cache/runtime factories, consumer adapters, and computeProcessingConfigFingerprint. The default preset uses the server-default provider bundle, matching the reference application. The createEngine methods remain aliases for reference consumers.

Provider credentials must be injected. Only named FromEnv adapter helpers read the environment when called. Classifier error observers are optional and cannot change classification results. Fingerprints include secret presence, never secret values; plugin overrides affect runtime fingerprints. The frozen reference default with caching disabled is processing-v1:8950483d4d7ec82b080b7f638bfaa0cc2aa6d6298e442c38a68a2e16c0d84b1a.

Applications own persistence, queueing, retry policy, and Effect integration.

The reference HEAD configuration hashes to d8e665…; the fingerprint helper aliases only that exact digest to the required 895048… base namespace. Changed configurations retain their own hashes.