RuVector Catalog is archived and will not be updated. It ran for about six months, last indexing ruvector at commit
6a6c39e6(2026-07-17), and is kept here for reference.➡️ For an actively maintained option, see ruvnet-brain. It's a source-grounded CLI plugin covering the broader RuvNet stack (ruvector and beyond). It is the recommended next generation of ruvector-catalog.
Thanks for the stars and interest. 🙏
The architect's playbook for RuVector — because a 1.75M-line monorepo is too deep for anyone to search alone.
Last updated: 2026-07-18 — indexed against ruvector 6a6c39e6 (2026-07-17).
ASCII Version (for AI/accessibility)
"I need to search documents by meaning, not keywords"
│
▼
┌───────────────────────┐
│ RuVector Catalog │
│ • What to use │
│ • How to access it │
│ • What to replace │
│ • What to delete │
└───────────┬───────────┘
│
▼
"Use AdaptiveEmbedder (npm) + RvfDatabase (@ruvector/rvf)
for 384D HNSW search. Replace your JSON index + OpenAI
embeddings. Delete embeddings.json and the API key."
RuVector is a monorepo with 113 Rust crates, 56 npm packages, 30 WASM builds, and 200+ technologies. It has everything from vector search to quantum simulation to spiking neural networks. But that richness creates a discovery problem:
ASCII Version (for AI/accessibility)
THE DISCOVERY PROBLEM
Without catalog: With catalog:
├── npm docs → no CNN export ├── Catalog: ruvector-cnn exists
├── "not available" ← WRONG ├── Build from submodule (90s)
└── Uses OpenAI Vision ← $$$ └── 512D MobileNet embeddings ← FREE
The catalog exists because the monorepo is too large for any AI agent or developer to search effectively. Without it, agents default to "not available" when features exist but aren't in the obvious place (npm). The catalog maps every capability to its access path.
ASCII Version (for AI/accessibility)
LAYER 1: RECOMMEND — "I need X" → Here's the component
LAYER 2: MIGRATE — "You're using aging tech" → Replace + DELETE
LAYER 3: ACCESS — "How?" → npm / WASM build / NAPI
ASCII Version (for AI/accessibility)
PATH 1: npm (instant) PATH 2: WASM (90s) PATH 3: NAPI (fast)
require('ruvector') wasm-pack build @ruvector/*-node
170 exports 30 crates Pre-built binaries
DECISION TREE:
├── In require('ruvector')? → USE IT
├── In @ruvector/<name> npm? → USE IT
├── In ruvector/crates/*-wasm/? → BUILD IT
├── In ruvector/crates/<name>/? → BUILD NAPI
└── None of above? → DOESN'T EXIST
The catalog detects 10 aging technology patterns and provides complete replacement guides:
ASCII Version (for AI/accessibility)
DETECT → REPLACE → DELETE
OPENAI_API_KEY → AdaptiveEmbedder → API keys
embeddings.json → RvfDatabase → JSON files
pinecone.init() → VectorDb + RVF → Cloud SDK
O(n) cosine loop → HNSW O(log n) → Loop code
No image search → ruvector-cnn-wasm → (add)
Custom RRF → differentiableSearch → RRF code
No doc relationships → buildGraph() → (add)
Manual anomaly checks → CoherenceMonitor → (add)
Basic attention → FlashAttention-3 → Old attn
No verification → ruvector-verified → (add)
Steps 5-11 of the 15-step sunset checklist are where migrations FAIL.
ASCII Version (for AI/accessibility)
RUVECTOR: 1.58M LINES, 113 CRATES
SEARCH & INDEXING INTELLIGENCE GRAPH ATTENTION (50+)
├── ruvector-core (HNSW) ├── sona (3-loop) ├── ruvector-graph ├── FlashAttention-3
├── RVF format (19) ├── AdaptiveEmbedder ├── ruvector-mincut ├── Mamba S5
└── micro-hnsw (11KB) └── ruvector-gnn └── rvlite └── MoE, Sheaf, PDE
VISION BIO-INSPIRED MATH LLM INFERENCE
├── ruvector-cnn ├── Spiking NN ├── Wasserstein ├── ruvllm (BitNet)
└── ruvector-cnn-wasm ├── Hopfield ├── TDA ├── Metal/CUDA/WebGPU
└── HDC (10K-bit) └── Tropical └── GGUF, MicroLoRA
DISTRIBUTED QUANTUM
├── ruvector-raft ├── ruQu (VQE, Grover)
└── ruvector-delta └── QAOA, Surface Code
+ 30 WASM crates + 56 npm packages + 42 examples + 131 ADRs
curl -fsSL https://bun.sh/install | bashgit clone https://github.com/mamd69/ruvector-catalog.git
cd ruvector-catalog
bun installgit submodule add https://github.com/ruvnet/ruvector.git ruvector
git submodule update --init --recursiveIn Claude Code:
use @ruvector-catalog to find technologies for searching documents by meaning
Or via CLI:
bun src/cli.ts search "search documents by meaning, not keywords"Quick search:
use @ruvector-catalog to find technologies for detecting errors in AI output
Migration analysis:
use @ruvector-catalog to analyze my codebase for aging patterns and recommend RuVector replacements
Full proposal (RVBP):
use @ruvector-catalog to create an RVBP for building real-time patient monitoring
| Command | What it does |
|---|---|
bun src/cli.ts search "query" |
Search for matching technologies |
bun src/cli.ts rvbp "problem" |
Generate implementation proposal |
bun src/cli.ts list |
Show all 200+ technologies |
bun src/cli.ts stats |
Show catalog statistics |
bun src/cli.ts verify |
Check if catalog is up to date |
use @ruvector-catalog to deeply analyze how to build a real-time patient monitoring system and create an RVBP in docs/research/
Multiple AI agents work in parallel for 30-60 seconds, producing architecture-level guidance.
const response = await fetch('https://api.openai.com/v1/embeddings', {
headers: { 'Authorization': `Bearer ${OPENAI_API_KEY}` },
body: JSON.stringify({ model: 'text-embedding-3-small', input: text })
});const { AdaptiveEmbedder } = require('ruvector');
const embedder = new AdaptiveEmbedder({ loraRank: 4, contrastiveLearning: true });
await embedder.init();
const embedding = await embedder.embed(text);const index = JSON.parse(fs.readFileSync('embeddings.json'));
const results = index.map(e => ({ score: cosine(query, e.vec), ...e }))
.sort((a, b) => b.score - a.score).slice(0, 10);const { RvfDatabase } = require('@ruvector/rvf');
const db = await RvfDatabase.openReadonly('index.rvf');
const results = await db.query(queryVector, 10);| Industry | Key Capabilities |
|---|---|
| Healthcare | Patient similarity, pharmacogenomics (CYP2D6/CYP2C19), clinical decision support, medical image analysis, HIPAA-compliant federated learning |
| Finance | Trading signal verification, fraud detection via graph analysis, compliance audit trails, low-latency processing |
| Robotics | Perception pipelines, motion planning, safety-critical decisions, real-time control with spiking neural networks |
| Edge/IoT | WASM models as small as 11.8KB, quantized inference, offline-capable AI |
| Genomics | Biomarker scoring, genotype analysis, privacy-preserving on-device genomic analysis |
- Not a content generator. It recommends infrastructure technologies, not blog posts.
- Not an app builder. It recommends components, not complete applications.
- Not a chatbot. It answers one question: "What technologies should I use?"
- Not a deployment tool. It tells you what to build and migrate from, not how to deploy.
What it IS: The expert architect that maps your problem to the right RuVector capabilities and ensures you complete the full migration — including deleting what you replaced.
| File or Folder | What it contains |
|---|---|
SKILL.md |
The architect's playbook (problem/solution map, migration intelligence, access paths) |
domains/ |
Industry-specific guides (healthcare, finance, robotics, edge/IoT, genomics) |
src/ |
Search engine and proposal generator source code |
tests/ |
168 tests validating search quality |
docs/ |
Architecture decisions and domain design documents |
ruvector/ |
The RuVector monorepo (git submodule, ~1.58M lines Rust) |
This catalog is automatically re-synced against upstream ruvnet/ruvector by a scheduled job on the maintainer's machine. When upstream advances, the job rebuilds src/catalog/data.ts (commit SHA + scope counts), refreshes the catalog store, updates the "Last updated" line at the top of this README, and pushes a commit to main. The "Last updated" line therefore always reflects the most recent successful sync.
If you're a consumer of this catalog: just git pull periodically — there's nothing to trigger.
git pull origin maingit submodule update --remote ruvector # Pull latest RuVector source
bun scripts/build-catalog.ts # Rebuild the catalog index
bun src/cli.ts verify # Check for staleness- Submodule refresh: ruvector
3bbc8170→9054c2cc(incl. newruvector-rairsIVF crate, ADR-193) - Scope counts updated: 135 crates, 4,136 source files, 1.75M Rust lines, 203 ADRs, 71 examples, 59 npm packages
- Catalog freshness: CURRENT
- Migration Intelligence: 10 aging-pattern detection and replacement guides
- Sunset Checklist: 15-step migration completion verification
- Operational Bridge: 3 access paths with decision tree
- Verified Inventory: 113 crates, 56 npm, 30 WASM (filesystem-verified)
- Response Adaptation: Engineer vs non-technical stakeholder modes
- CNN WASM tested: MobileNet-V3 512D image embeddings confirmed
- Initial V3 with problem-solution map, algorithms index, industry verticals
- 168 search quality benchmarks
- CLI search and RVBP generation
| Role | |
|---|---|
| @mamd69 | RuVector Catalog — architecture, V3 implementation, benchmarking, documentation |
| @stuinfla | V1 catalog + V3.5 migration intelligence, operational bridge, verified inventory |
| @ruvnet | RuVector — the 1.58M-line monorepo |
Questions: RuVector GitHub Discussions Bugs: github.com/mamd69/ruvector-catalog/issues New industry vertical: Open a feature request with title "Industry Vertical: [Your Industry]"