Build with the Soulcraft platform.
Two products power every Soulcraft app and every external integration — Brainy, the open-source knowledge graph, and Cortex, the native Rust accelerator that scales it to billions on commodity hardware. Same code. Same files. Same API.
Brainy
MIT · Open Source · v7.30The open-source knowledge operating system.
Vector search, graph relationships, and document storage unified in one library. Self-hosted, free forever, zero config — runs on its own up to millions of entities.
Cortex
Native Rust · v3.0 coming Q3 202610B+ even on a laptop. 100B+ on a server.
The native Rust accelerator that scales Brainy from millions to billions. DiskANN algorithm (Microsoft Research, 2019), reimplemented in 100% Rust. Same Brainy code, same files, same API — scale by adding servers, no managed service in the middle.
Jump into Brainy
Brainy is where most developers start. It's MIT-licensed, runs on its own, and ships with everything you need to build a real knowledge graph.
Plain-language explanation of what Brainy does and why it exists. No jargon, no code.
Install Brainy and write your first knowledge query.
Triple intelligence, noun types, verb types, and zero-config design.
Storage adapters, the Find system, batch operations, and plugins.
Every method — add, find, relate, branch, version, VFS, neural.
Rust-powered SIMD, quantization, and native embeddings for production.
What is Brainy?
Brainy is a knowledge operating system that combines three types of data intelligence into one simple API:
- Vector Search — Find content by meaning, not just keywords
- Graph Relationships — Connect entities with typed relationships (127 verb types)
- Document Filtering — Query metadata with MongoDB-style operators
This combination — called Triple Intelligence — lets you replace multiple specialized databases (Pinecone + Neo4j + MongoDB) with a single unified system.
Quick Example
import { Brainy, NounType, VerbType } from '@soulcraft/brainy'
const brain = new Brainy()
await brain.init() // zero config — auto-detects everything
const reactId = await brain.add({
data: "React — a JavaScript library for building user interfaces",
type: NounType.Concept,
metadata: { category: "frontend", year: 2013 }
})
// Triple Intelligence: vector + graph + metadata in one query
const results = await brain.find({
query: "modern frontend frameworks", // vector similarity
where: { year: { greaterThan: 2015 } }, // metadata filter
connected: { to: reactId, depth: 2 } // graph traversal
})Key Brainy Features
Zero Config
Auto-detects storage, embeddings, and indexes. Just install and use.
Git-Style Versioning
Fork, commit, and time-travel your data. No other vector DB has this.
Native Performance
Rust-powered HNSW, SIMD, and GPU acceleration on Node.js 22+ and Bun 1.3+.
MIT Licensed
Free forever. No vendor lock-in. Self-host or use our cloud.
Jump into Cortex
Cortex is the native Rust accelerator. Same Brainy code, same files, same API — orders of magnitude faster on vector search, aggregation, and large graph traversals at billion-scale.
The architecture behind 10B-entity workloads on a single commodity box — mmap NVMe, DiskANN, LSM metadata.
Vamana α-pruned graph + Product Quantization, reimplemented from the paper in 100% pure Rust.
How the open-core upgrade story compares to managed-only and cluster-required alternatives.
Where Cortex spends time and memory, what it accelerates, and what stays on the JS path.
Native group-by, count, and sortTopK across billion-row metadata indexes.
Why Cortex 3.0 widens entity IDs from u32 to u64 — and what that unlocks at 10B-scale.
Ready to build?