Hunter-RAG

Hunter-RAG is a retrieval substrate. Where most systems cut a document into arbitrary chunks, it works in coherent document units — structure-preserving pieces an agent can name, locate, and operate on directly. Retrieval becomes active rather than approximate: the system returns not only relevant text but the path to it and the evidence behind it.

A language model is only as trustworthy as what it retrieves, and most retrieval fails quietly — the right passage is found but unusable, or an answer reads fluently yet cannot be traced to a source. In domains where being wrong is expensive, an answer without its evidence is not an answer.

Arc treats retrieval as a substrate to own, not a library to wire together. The core mechanism is protected; the public claim is narrower and sufficient — that grounding, traceability, and addressable structure belong in the retrieval layer itself, not bolted on afterwards. Hunter-RAG is the foundation several Arc systems are built on.

Productisation