A vector store that lives inside your process.

A lightweight, lightning-fast, in-process vector database.

pip install zvec Read the docs →

network hops

dependency, embedded

~µs

in-memory lookup

Metric labels are illustrative of the in-process design, not benchmarks.

rag.py

import zvec

db = zvec.Index(dim=768)
db.add(ids, embeddings)   # no server

hits = db.search(
    query_vec,
    k=5,
)
for id, score in hits:
    print(id, score)

No server to run

Embedded in your application process — no daemon, no ports, no ops. Just import and index.

Lightweight by design

A small footprint meant to sit beside your model code, not a cluster to operate alongside it.

Built for speed

In-memory similarity search tuned to keep retrieval off the critical path of your app.