CoreCortexView source
Retrieval

Return ranked evidence, not opaque answers

CoreCortex's query contract makes retrieval decisions observable and leaves response synthesis to the calling application.

Request

httpCoreCortex Core
POST /v1/corpora/{corpusId}/query
Content-Type: application/json

{
  "query": "Which warnings mention liver injury?",
  "retrievalPlan": "evidence-search",
  "limit": 12
}

Response responsibilities

  • Ranked evidence with source and fragment identity
  • Provenance required to trace a result to the record
  • Retrieval-plan and index-version details for reproducibility
  • Scores and diagnostics suitable for evaluation

Why synthesis stays outside

Separating retrieval from generation lets teams evaluate memory quality without conflating it with model style or prompt behavior. Applications can choose their own model, citation format, and interaction pattern over the same evidence contract.

Lifecycle operations

The v1alpha1 API also defines configuration validation, corpus revisions, record upsert and deletion, corpus inspection, and durable job status. It is an evolving contract rather than a hosted public endpoint today.

Continue readingSee the architecture→