Biblicus

Make your documents usable by your assistant, then decide later how you will search and retrieve them.

Embedding index (in-memory)

This backend builds an embedding index in memory and queries it using exact cosine similarity.

It is intended for textbook demos and small corpora where you want a “real” embedding retrieval loop without running an external vector database.

Backend ID

embedding-index-inmemory

What it builds

This backend builds a retrieval snapshot that materializes:

All of this lives in memory while the process is running. For safety, the backend enforces explicit caps so a build does not accidentally consume unbounded memory.

Chunking

Embeddings are computed over chunks. Chunking is configured per configuration by selecting a chunker and its configuration.

Chunking is part of the index contract: evidence references chunk boundaries so you can trace retrieval outputs back to the original item text.

Dependencies

This backend does not require a database or server.