Enjab RAG
Version history for Enjab RAG, the shared knowledge base. Each entry says what changed and the action (if any) to adopt it.
Changes to Enjab RAG, newest first. Enjab RAG is a hosted service, most changes are server-side and need no action from connected tools. To connect or re-check a tool, see Connect to Enjab RAG.
2026.06.11a: Enjab RAG is live
Enjab RAG · 11 Jun 2026
New. The shared knowledge brain for Enjab's agents is live at rag.enjab.ae. It ingests files, text, images, audio, and links; extracts and processes them (OCR, transcription, then an auto title, summary, tags, and structured metadata); and serves retrieval over a native MCP server (search + fetch tools) and a REST API (/api/search, /api/ingest). Embeddings use OpenAI text-embedding-3-large (3072-dim) in Postgres + pgvector.
Apply. Connect every Enjab agent to it so they share one source of operational knowledge, see Connect an agent. Mint a key at rag.enjab.ae/dash/keys.
2026.06.11b: Smart chunking (structure-aware + Contextual Retrieval)
Enjab RAG · 11 Jun 2026
New. Chunking now respects document structure (tables keep their header row on every chunk, prose keeps its section heading) and applies Contextual Retrieval: a model writes a one-line, document-aware context for each chunk that is embedded alongside it, so short or ambiguous chunks (a spreadsheet row, a bare figure) retrieve correctly. Large boundary-less documents (big spreadsheets) that previously failed to ingest now chunk and index reliably.
Apply. Nothing, server-side. Retrieval quality improves automatically; re-ingest a document if you want it re-chunked under the new logic.