GrainVDB combines hardware ARM NEON SIMD vectorization with 2D Metal GPU compute shaders over Apple's 800 GB/s Unified Memory (UMA). Sub-millisecond local retrieval with zero PCIe transfer penalties and zero recurring cloud bills.
Experience real-time local semantic retrieval, topic predicate filtering, and hallucination / topology fracture auditing directly in your browser.
Measured on real Apple Silicon (Darwin arm64) using exact scan, batch compute, and memory-mapped indexes.
| Engine / Mode | Latency (p50) | Throughput | Recall |
|---|---|---|---|
| ARM NEON Fast-Path | 0.108 ms | ~9,250 QPS | 100% |
| Metal GPU 2D Grid | 1.906 ms | 734 QPS | 100% |
| HNSW Graph Index | 0.735 ms | Sub-linear | 42% - 95% |
| Page-Aligned mmap | 0.54 ms | Instant Open | N/A |
Recurring cloud bills + bandwidth ingress/egress fees + network latency.
Zero recurring API bills. Complete data privacy. Runs directly on your existing hardware.
First-class packages for Swift Package Manager, Python, and LangChain.
import GrainVDB
// 1. Initialize embedded vector DB
let db = try GrainVDB(dimension: 128, mode: .exact)
// 2. Add embeddings
try db.addVectors(embeddingMatrix)
// 3. Sub-millisecond retrieval with metadata closure filter
let results = try db.search(query: queryVector, k: 5) { id, meta in
meta?["category"] as? String == "Privacy"
}
for doc in results {
print("Match: \(doc.id), Score: \(doc.score)")
}from grainvdb.integrations import GrainVDBVectorStore
from langchain_community.embeddings import OllamaEmbeddings
# Initialize local embedding model & GrainVDB
embeddings = OllamaEmbeddings(model="nomic-embed-text")
vectorstore = GrainVDBVectorStore(embedding=embeddings, dim=768)
# Ingest and query in 0.1ms
vectorstore.add_texts([
"Apple M2 Ultra features 800 GB/s unified memory.",
"GrainVDB runs zero-copy vector search on Mac."
])
docs = vectorstore.similarity_search("memory bandwidth", k=1)
print(docs[0].page_content)Free for open-source. Simple, fair commercial licenses for proprietary apps and enterprise deployments.
Free forever for personal, academic & open-source use.
Perpetual rights to embed GrainVDB in proprietary apps.
For law firms, healthcare, media studios & Mac clusters.