Skip to content

Changelog

Initial release of the RAG API.

  • Knowledge Base CRUD — Create, list, get, update, and delete knowledge bases
  • Document Upload — Multipart file upload with 16 supported file types (9 Tier 1 parsed, 7 Tier 2 direct)
  • Text Ingestion — Ingest raw text via JSON body, same pipeline as file uploads
  • Document Replacement — Full re-process via PUT
  • Search — Natural language search with Voyage AI embeddings + cross-encoder reranking
  • Metadata Filtering — Filter search results by document metadata ($eq, $ne, $gt, $gte, $lt, $lte, $in, $nin)
  • Chunk Listing — Inspect how documents were segmented, cursor-paginated
  • Usage Tracking — Monitor pages processed and search queries consumed
  • Rate Limiting — Per-plan rate limits with clear error responses
  • SDKs — TypeScript and Python SDKs with full type coverage
  • Embedding model: Voyage AI voyage-4 (1024 dimensions)
  • Reranking model: Voyage AI rerank-2.5
  • Benchmark: nDCG@10 = 0.94 on SciFact dataset