Documents, RAG & visual search

Chat with your own files locally — parsing, embeddings, visual document retrieval, rerank, and citations on-device.

Bedrova can chat with your own documents — entirely on your Mac.

Screenshot: a document chat with cited sources — added at launch.

Import & parse

Add PDF, DOCX, TXT, Markdown, HTML, CSV and JSON. Bedrova parses them locally with its bundled parser, and scanned pages go through on-device OCR. Heavier document layout and table extraction is available as an optional advanced parser pack if you need it. Either way your files never leave the machine.

Index

Documents are chunked with stable source spans and embedded with your chosen local embedding/search model into a local vector store. Scanned PDFs and visual documents can also be indexed by page image so retrieval does not depend only on OCR text. When a document fits the model’s context, Bedrova can skip retrieval and use the whole thing.

Ask

Retrieval uses hybrid search, visual document retrieval where available, and reranking before packing the best context into the prompt. Answers show citations you can click to jump to the source page, span, image, or timestamp. Rebuild, repair, cancel, or delete an index any time.

Scope

This is personal, local doc/folder/media search. Durable, team, or agentic knowledge bases are a job for Packwolf — Bedrova stays personal and on-device.

API

Manage documents and query retrieval via the native /api/v1/documents/* and /api/v1/rag/query routes. Multimodal search and rerank use /api/v1/search and /api/v1/rerank.