# KoutenDB

A locality-first document and vector database that lets applications place related data into explicit ring coordinates and retrieve the bounded nearby working set before ranking, filtering, or building LLM context, reducing the data scanned per query.

KoutenDB is a document and vector database built around the idea that when an application already knows which data is related -- a tenant, a repository, a user -- it should not rank, transfer, or scan the rest. Data is placed into ring coordinates, and a query retrieves the bounded nearby working set as a first-stage retrieval boundary rather than a filter applied after a broad search.

Related records can be stored as nearby subrings and read back as one bounded bundle with per-subring limits and sort order, letting an endpoint state the shape it needs directly. The project's benchmarks show large reductions in scanned records, candidate memory, and estimated tokens per query on RAG and general workloads when ring selection is used.

KoutenDB targets RAG and AI retrieval as well as tenant-scoped web systems, user profile pages, and other services where related data is known ahead of a read.

Website: https://puffball1567.github.io/koutendb/