Sextant

Navigate, don’t trawl.

Sextant is an SSD-resident vector index engine: billion-scale vector search on a single commodity machine. Cheap measurements instead of exhaustive scans — so latency stays low, RAM stays small, and the bill stays flat.

corpus · billion vectors on NVMe the target query measured angle = route
Routing plane

Search a billion vectors without reading a billion vectors

PCA-projected signatures route each query to a handful of leaf extents before anything is read off disk. Probe effort is measured in NVMe pages and grows far slower than the corpus — sublinear by contract, not by caching luck.

Build

Build where you are, at the scale you need

The streaming tree build keeps memory bounded by chunk size, not corpus size. No memory arms race, no build cluster — the finished index lands as one compact file on ordinary NVMe.

Recall

Vectors at a few hundred bytes, answers at full quality

Per-leaf quantizers compress each vector to a few hundred bytes; exact rerank against the originals claws back the compression. Users see near-perfect results — the hardware budget never notices.

Freshness

Your data changes; your index keeps up

Append, delete, vacuum, defrag — Sextant is a living store, not a frozen artifact. Native predicate filters return the nearest result that actually qualifies, not just the nearest one.

Storage

Predictable performance, near-zero idle cost

Sextant owns its storage stack — direct I/O plus a built-in W-TinyLFU cache. Hot queries serve at SSD speed; idle RAM stays near zero. No page-cache tuning, no terabytes of memory reserved for data at rest.

Status · pre-release sextant is under active development. there is no release yet — no downloads, no benchmark claims, no blog posts. when there is, it will be here.

Why the name

A sextant doesn’t find the star. It measures the angle to it — and from that measurement fixes your position. Here, the star is the vector you’re looking for, the navigator is your query, and the sextant is the routing plane: cheap measurements first, expensive chart-reading only where it matters.

You carry the instrument, not the ocean’s chart. The routing table lives in DRAM; the billion-vector corpus stays on disk.