Skip to content

openai-vector-store

Reads OpenAI vector stores through the real openai SDK ($OPENAI_API_KEY, --base-url, --organization, --vector-store repeatable, default every store). Endpoints used: GET /v1/vector_stores, .../{id}/files, .../{id}/files/{file_id}/content, GET /v1/files/{file_id}.

  • OpenAI does not expose embeddings, so records are the parsed text chunks, one per content part (<file_id>#<n>), and HMS encodes them.
  • Namespace = vector store id (--namespace-by name for its name). Metadata: file attributes flat, file_id, filename, vector_store_id, vector_store_name, chunk_index, store metadata. created_at = the file's creation time.
  • A file that is not completed, or whose content has more pages than the API lets a client follow, is rejected visibly.
  • Verification: MOCK. The real SDK runs against a mock of the documented API (https://platform.openai.com/docs/api-reference/vector-stores-files); no live OpenAI call was made.