LynseDB

LynseDB is a high-performance, low-cost vector database with a Python-first API and a Rust storage/search backend. It can run embedded inside a single Python process or as an HTTP server for multi-process and production deployments.

The API is intentionally small: create a database, create a collection, write vectors with metadata, build an index when you need faster search, and query the same collection locally or remotely with the same high-level Python calls.

Use it when you want vector search that starts as an embedded local library, can move to a service when your app grows, and does not require a heavy database cluster for common semantic search, RAG, agent memory, and multimodal retrieval workloads.

What LynseDB is good at

• Local development, notebooks, scripts, tests, and small services that want an embedded vector store.
• Remote deployments where several workers or services need to share one database process.
• Low-operational-cost deployments where predictable resource use matters.
• Standard SQL-style metadata filtering with where expressions.
• Multiple vector representations per record through named vector fields.
• Dense, sparse, text, and hybrid retrieval.
• Recall-first search with explicit index and search knobs.

Search more than embeddings

LynseDB provides native distance paths for domain data as well as embedding vectors. They use the same IDs, metadata filters, persistence, and local or remote client API.

Data	Metrics	Example workloads
Embeddings	inner product, squared L2, cosine	RAG, semantic and multimodal retrieval
Numeric features	Manhattan/L1, Chebyshev, Canberra	anomaly matching, tolerances, sensor and tabular features
Coordinates	Haversine in meters	nearby POI, fleet and device search
Binary fingerprints	Hamming, Jaccard/Tanimoto, Sørensen-Dice	molecular fingerprints, deduplication, genomic sketches
Aligned profiles	Pearson correlation distance	sensor curves, behavior profiles, gene expression
Distributions and abundance	Hellinger, Jensen–Shannon, Wasserstein-1D, Bray–Curtis	model drift, topics, forecasts, histograms and ecology

Exact Flat search supports the complete suite. HNSW supports L1, Haversine, correlation, Hellinger, Wasserstein-1D, Jensen–Shannon, and Chebyshev, while binary Flat search uses a one-bit-per-dimension hot representation for low memory bandwidth. Read Domain-aware distance metrics for input contracts and index compatibility.

Execution modes

Mode	How to start	Best for	Notes
Local	lynse.VectorDBClient() or lynse.VectorDBClient(uri="./data")	notebooks, scripts, single-process apps	No server and no network hop. Do not share the same local path between independent processes.
Remote	lynse serve ... plus lynse.VectorDBClient("http://host:7637")	web services, workers, production	Use HTTP mode when more than one process writes or searches the same data.

Install

pip install lynsedb

Python 3.9 or newer is required. Wheels are published for common macOS, Linux, and Linux environments. Native Windows environments are not supported; on Windows, use WSL 2 (Windows Subsystem for Linux) or Docker.

First search

import numpy as np
import lynse

client = lynse.VectorDBClient(uri="./lynsedb-data")
db = client.create_database("demo", drop_if_exists=True)
collection = db.require_collection("documents", dim=4, drop_if_exists=True)

collection.add(
    ids=["intro", "guide", "notes-fr"],
    vectors=[
        [0.10, 0.20, 0.30, 0.40],
        [0.11, 0.19, 0.29, 0.39],
        [0.80, 0.10, 0.20, 0.10],
    ],
    fields=[
        {"title": "intro", "lang": "en"},
        {"title": "guide", "lang": "en"},
        {"title": "notes", "lang": "fr"},
    ],
)

collection.build_index("FLAT-L2")

result = collection.search(
    np.array([0.10, 0.20, 0.30, 0.40], dtype=np.float32),
    k=2,
    where="lang = 'en'",
    return_fields=True,
)

print(result.to_list())

Learn next

• Start with the Learning path if you want a step-by-step curriculum from zero to production.
• Use the Quickstart for the shortest complete write, index, search, query, and delete lifecycle.
• Read Core concepts to understand clients, databases, collections, IDs, fields, indexes, and result objects.
• Use Connect and deploy to choose local or remote mode and configure the server.
• Use Search and filter and the Metadata filter cookbook for vector search, standard SQL-style filters, BM25 search, hybrid search, and reranking.
• Use Indexing guide and Performance tuning to choose between flat, HNSW, IVF, SPANN, DiskANN, and quantized indexes with explicit metric suffixes.
• Use Domain-aware distance metrics for coordinates, binary fingerprints, aligned profiles, and distributions.
• Use Build a RAG workflow for an end-to-end retrieval example.

Stability notes

LynseDB is still evolving. Pin versions for deployments and test migrations before upgrading. For concurrent production access, prefer the HTTP server over sharing one local data directory across independent Python processes.