ELIPS lives inside your process. It is to vector search what SQLite is to relational data.

From docs/overview/introduction.md: ELIPS is designed for CLIs, desktop apps, notebooks, edge workloads, single-process APIs, and tests — anything that wants local, in-process retrieval without operating a separate service.

Why a vector DB exists: text → embedding → nearest-neighbor lookup over a billion-row corpus in milliseconds.

What you actually get

ANN and exact vector search over the same vault
First-class documents with chunk coordinates and embedding lineage
Native text-first and hybrid query APIs
Metadata filters with planner-side candidate narrowing
WAL recovery plus segmented persistence
Shared read-only mode for multi-reader serving
Python and C++ SDKs over the same C++23 core

What ELIPS is not

Not a distributed database. No replication protocol, no sharding service.
Not a SaaS. Nothing to deploy. Your process opens a directory.
Not a vector-only store. Documents, payloads, lineage are first-class.