Self-hosted RAG with CDOC: chunking, metadata filters, reranking
If your corpus fits a cache, you don't need a separate vector database to do retrieval. CDOC adds documents with auto-chunking, filtered search, and reranking, all local.
Retrieval-augmented generation usually means standing up a second system: a vector database, an embedding pipeline, a chunker, a reranker. For a corpus that fits a cache, that's a lot of moving parts for a common case. CDOC folds it into Crowkis.
CDOC ADD policy-2026 "<long text>" CHUNK 512 OVERLAP 64 META team=legal CDOC SEARCH "how long is the refund window?" K 3 FILTER team=legal RERANK
Documents are auto-chunked into overlapping passages with metadata you attach. Search runs a filtered approximate-nearest-neighbour query, so a tenant's documents are never starved, with an optional rerank pass using the local cross-encoder. Results come back as id, text, and score, the shape every RAG pipeline expects.
Because CDOC shares the bundled local embedder, retrieval inherits the same zero-egress property. That matters when 'the documents' are contracts, tickets, or anything you can't ship to a hosted API. CDOC isn't trying to replace a large-scale vector database, it's trying to remove one from the diagram for the many apps that don't need that scale.