How does HNSW vector search work client-side?

Vector Studio implements Hierarchical Navigable Small World (HNSW) multi-layer graphs directly in JavaScript, routing vector queries with logarithmic O(log N) search latency.

Does my vector data or embedding leave my browser?

No. All embeddings, clustering, distance calculations, and graph traversals execute 100% in-browser on Web Workers and TypedArrays with zero cloud egress.

What checkout methods are supported?

Credit cards, Apple Pay, and Google Pay via Stripe Checkout, plus instant Solana Web3 settlement via DePay for $19 lifetime access.

Vector Studio | HNSW Vector Search Engine & Hybrid Lab

Viewport Controls

2D / 3D Random Projection

Projection Mode

Active Points Count

Color By Metadata

AI / NLP

Database

Physics

Query Parameters

Real-time vector & metadata search

Query Text or Vector Array

Top-K Neighbors 5

Metadata Filter (JSON)

Ranked Results Deck

0 results (0.0 ms)

Enter a query or click "Run Search" to simulate similarity search.

🔒 SIMILARITY RESULTS LOCKED

Ranked nearest neighbors, distance vectors, and similarity scores are blurred on free tier.

Unlock Pro Access ($19)

HNSW Graph Hierarchy

Skip-graph layers & highway routing

Layer 0 (Base)

Layer 1 (Express)

Layer 2 (Highway)

Graph Structural Diagnostics

Algorithm configuration and edge density

Total Layers 3

Avg Links / Node 14.2

Entry Point ID doc_42

ef_construction 64

Greedy Traversal Log

HNSW Ready. Click 'Step-Through Greedy Routing' to trace search path.

Hybrid Search & Fusion Tuning

Dense Vector + Okapi BM25 Lexical Ranking

Hybrid Query (Text + Vector)

Fusion Strategy

Dense vs Sparse Alpha Weight (α) 0.50

0.0 (Pure BM25 Sparse) 0.5 (Balanced Hybrid) 1.0 (Pure Dense HNSW)

1. Dense HNSW Rank

2. Sparse BM25 Rank

3. Blended Fusion Rank

⚖️ HYBRID FUSION BLURRED

Reciprocal Rank Fusion outputs and blended rankings require Pro active status.

Unlock Pro Access ($19)

Synthetic Benchmark Config

Measure QPS, Latency & Recall@K

Vector Count

Dimension

Queries Count

Comparative Performance Metrics

Ready to execute.

HNSW QPS -

Flat Scan QPS -

Recall @ K -

Speedup Factor -

Query Latency Comparison (Average ms)

HNSW Graph

0.0 ms

Flat Brute Force

0.0 ms

Detailed Execution Log

Click 'Run Benchmark Suite' to generate vectors and execute tests.

Insert Vector Document

Add items with text and metadata

Document ID

Document Text

Vector Values (comma-separated or empty to auto-generate)

Metadata (JSON)

Collection Overview

Summary statistics

Select a collection to view statistics.

Cluster Controls

K-Means++ & Partitioning

Algorithm

Clusters (K)

Max Iterations

Random Seed

Cohesion & Silhouette Quality

Ready to analyze collection clusters

Total Clusters -

Silhouette Score -

Inertia (WCSS) -

Computation Time -

Discovered Cluster Partitions & Medoids

Exemplars, dispersion, and metadata distributions

Click 'Run Cluster Analysis' to partition vectors and discover representative exemplars.

Upgrade to Vector Studio Pro