[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-10910":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":9,"totalLinesOfCode":9,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":9,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":9,"pushedAt":9,"updatedAt":42,"readmeContent":43,"aiSummary":44,"trendingCount":16,"starSnapshotCount":16,"syncStatus":45,"lastSyncTime":46,"discoverSource":47},10910,"RuVector","ruvnet\u002FRuVector","ruvnet","RuVector is a High Performance, Real-Time, Self-Learning Ai, Vector GNN, Memory DB built in Rust.",null,"https:\u002F\u002Fgithub.com\u002Fruvnet\u002FRuVector","Rust",4220,558,33,35,0,19,49,188,57,30.24,false,"main",[25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41],"ai","ai-ocr","gnn","graph","llm-inference","low-latency","neo4j","ocr","onnx","rust","vector","wasm","attention-mechanism","gnn-model","gnns","graph-neural-networks","mincut","2026-06-12 02:02:28","# RuVector — A Self-Learning, Vector Memory & Agentic Operating System\n[![CES 2026 Innovation Award](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🏅_CES_2026-Innovation_Award-gold.svg)](https:\u002F\u002Fcognitum.one)\n[![GitHub Trending](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🔥_GitHub-Trending-orange.svg)](https:\u002F\u002Fgithub.com\u002Fruvnet\u002Fruvector)\n\n[![Crates.io](https:\u002F\u002Fimg.shields.io\u002Fcrates\u002Fv\u002Fruvector-core.svg)](https:\u002F\u002Fcrates.io\u002Fcrates\u002Fruvector-core)\n[![npm](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002Fruvector.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fruvector)\n[![Downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdt\u002Fruvector.svg?label=Downloads)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fruvector)\n[![Monthly Downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdm\u002Fruvector.svg?label=Monthly%20Downloads)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fruvector)\n[![ruv.io](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fruv.io-website-purple.svg)](https:\u002F\u002Fruv.io)\n[![MIT License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-blue.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT)\n\n### **The self-learning, self-optimizing vector database — with graph intelligence, local AI, and PostgreSQL built in.**\n\n> Created by [rUv](https:\u002F\u002Fruv.io) and powering [Cognitum](https:\u002F\u002Fcognitum.one), a 🏅 **CES 2026 Innovation Awards Honoree** — the world's first Agentic Chip designed to be always running for AI agents. Tens of thousands of agents, near-zero power, learns from every signal. [Learn more →](https:\u002F\u002Fcognitum.one)\n\n\n```bash\nnpx ruvector\n```\n\n####  Most vector databases store your data and search it — the same way, every time. \n\n#### **RuVector** is fundamentally different. It watches how you use it and gets smarter: search results improve automatically, the system tunes itself to your workload, and it runs AI models right on your hardware — no cloud APIs, no per-query bills, GPUs optional, CPUs preferred. It drops into PostgreSQL, runs in browsers, and ships as a single file. \n\nOpen source. ❤️ Free forever.\n\n```\nUser Query → [SONA Engine] → Model Response → User Feedback\n                  ↑                                 │\n                  └─────── Learning Signal ─────────┘\n                         (\u003C 1ms adaptation)\n```\n\n\u003Cdetails>\n\u003Csummary>🔍 RuVector vs Typical Vector Databases (25 differences)\u003C\u002Fsummary>\n\n| | RuVector | Typical Vector DB |\n|---|---|---|\n| **Self-Learning & Optimization** | | |\n| [Search quality](.\u002Fcrates\u002Fruvector-gnn) | 🧠 GNN learns from every query — results improve over time | Static — same results every time |\n| [Self-optimizing](.\u002Fcrates\u002Fsona) | ⚡ SONA auto-tunes routing, ranking, and compression to your workload | Manual tuning required |\n| [50+ attention mechanisms](.\u002Fcrates\u002Fruvector-attention) | 🎯 FlashAttention-3, MLA, Mamba SSM, linear, graph, hyperbolic, [mincut-gated](.\u002Fcrates\u002Fruvector-attn-mincut) | Basic similarity only |\n| [Transfer learning](.\u002Fcrates\u002Fruvector-domain-expansion) | 🔄 Knowledge transfers across domains — new tasks bootstrap from past learning | Start from scratch each time |\n| **Search & Retrieval** | | |\n| [Hybrid search](.\u002Fcrates\u002Fruvector-core) | 🔍 Sparse vectors + dense vectors with RRF fusion — 20-49% better retrieval | Keyword OR vector, not both |\n| [Graph RAG](.\u002Fcrates\u002Fruvector-core) | 📊 Knowledge graph + community detection for multi-hop queries — 30-60% improvement | Naive chunk-based RAG |\n| [DiskANN](.\u002Fcrates\u002Fruvector-core) | 💾 Billion-scale SSD-backed ANN with \u003C10ms latency via Vamana graph | Memory-only indexes |\n| [TurboQuant](.\u002Fcrates\u002Fruvllm) | ⚡ 2-4 bit KV-cache quantization — 6-8x memory savings with \u003C0.5% quality loss | No quantization or 8-bit only |\n| [ColBERT multi-vector](.\u002Fcrates\u002Fruvector-core) | 🎯 Per-token late interaction retrieval (MaxSim) for fine-grained matching | Single-vector only |\n| [Matryoshka embeddings](.\u002Fcrates\u002Fruvector-core) | 🪆 Adaptive-dimension search — coarse-to-fine funnel for speed with minimal recall loss | Fixed dimensions only |\n| **Graph & Relationships** | | |\n| [Graph queries](.\u002Fcrates\u002Fruvector-graph) | 🔗 Full Cypher engine — `MATCH (a)-[:KNOWS]->(b)` like Neo4j | Flat list of results |\n| [Graph transformers](.\u002Fcrates\u002Fruvector-graph-transformer) | 🔬 8 verified modules: physics, bio, manifold, temporal, economic | No graph support |\n| [Hyperedges](.\u002Fcrates\u002Fruvector-graph) | 🕸️ Connect 3+ nodes at once — model group relationships natively | Pairwise only |\n| **AI & Compute** | | |\n| [Local LLMs](.\u002Fcrates\u002Fruvllm) | 🤖 Run models on your hardware — Metal, CUDA, WebGPU, no API costs | Cloud API required (pay per call) |\n| [Sublinear solvers](.\u002Fcrates\u002Fruvector-solver) | 📐 O(log n) PageRank, spectral methods, sparse linear systems | Not available |\n| [Graph sparsifier](.\u002Fcrates\u002Fruvector-sparsifier) | 🕸️ Keeps a small shadow graph that tracks the full graph's structure in real time | Not available |\n| [Genomics](.\u002Fexamples\u002Fdna) | 🧬 Variant calling, protein translation, HNSW k-mer search in 12 ms | Not available |\n| [Quantum coherence](.\u002Fcrates\u002Fruqu) | ⚛️ Error correction via dynamic min-cut optimization | Not available |\n| **Database & Platform** | | |\n| [PostgreSQL](.\u002Fcrates\u002Fruvector-postgres) | 🐘 230+ SQL functions — drop into your existing database, [pgvector replacement](.\u002Fdocs\u002Fpostgres\u002F) | Separate service to manage |\n| [Deploy anywhere](.\u002Fcrates\u002Frvf\u002FREADME.md) | 🌐 One file — servers, browsers, phones, IoT, bare metal, WASM (58 KB) | Cloud server required |\n| [Cognitive containers](.\u002Fcrates\u002Frvf\u002FREADME.md) | 🚀 Single `.rvf` file boots as a service in 125 ms — includes vectors, models, kernel | Configure a cluster |\n| [Live updates](.\u002Fcrates\u002Fruvector-core) | ⚡ Update vectors and graph connections instantly, no downtime | Rebuild index or wait |\n| **Operations** | | |\n| [Tamper-proof audit](.\u002Fcrates\u002Frvf\u002Frvf-crypto) | 🔐 Cryptographic witness chain records every operation automatically | Manual logging |\n| [Branch your data](.\u002Fcrates\u002Frvf\u002Frvf-cow) | 🌿 Git-like COW branching — 1M vectors, 100 edits = ~2.5 MB branch | Copy everything |\n| [Scale out](.\u002Fcrates\u002Fruvector-replication) | 📈 Raft consensus, multi-master replication, auto-sharding | Paid tiers, per-vector pricing |\n| [Post-quantum crypto](.\u002Fcrates\u002Frvf\u002Frvf-crypto) | 🛡️ ML-DSA-65 and Ed25519 signatures on every segment | Not available |\n| Cost | 💰 Free forever — open source (MIT) | Per-query or per-vector pricing |\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>📋 See Full Capabilities (75 features across 10 categories)\u003C\u002Fsummary>\n\n**Core Vector Database**\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 1 | [**Store vectors**](.\u002Fcrates\u002Fruvector-core) | Embeddings from OpenAI, Cohere, local ONNX with HNSW indexing and SIMD acceleration |\n| 2 | [**Query with Cypher**](.\u002Fcrates\u002Fruvector-graph) | Graph queries like Neo4j — `MATCH (a)-[:SIMILAR]->(b)` with hyperedges |\n| 3 | [**The index learns**](.\u002Fcrates\u002Fruvector-gnn) | GNN layers make search results improve over time — every query teaches the system |\n| 4 | [**Hyperbolic HNSW**](.\u002Fcrates\u002Fruvector-hyperbolic-hnsw) | Hierarchy-aware search in Poincare ball space — better for trees and taxonomies |\n| 5 | [**Compress automatically**](.\u002Fcrates\u002Fruvector-temporal-tensor) | 2-32x memory reduction with adaptive tiered compression and temporal tensor reuse |\n| 6 | [**Metadata filtering**](.\u002Fcrates\u002Fruvector-filter) | Filter search results by any field before scanning vectors — fast hybrid queries |\n| 7 | [**Collections**](.\u002Fcrates\u002Fruvector-collections) | Multi-tenant, schema-managed collections — isolate data per customer or project |\n| 8 | [**Snapshots**](.\u002Fcrates\u002Fruvector-snapshot) | Point-in-time backups — restore your database to any previous state |\n\n**Advanced Search & Retrieval** *(new in v2.1.0)*\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 8a | [**Hybrid search (RRF)**](.\u002Fcrates\u002Fruvector-core) | Sparse + dense vector fusion with Reciprocal Rank Fusion — 20-49% retrieval improvement |\n| 8b | [**Graph RAG**](.\u002Fcrates\u002Fruvector-core) | Knowledge graph + Leiden community detection + local\u002Fglobal\u002Fhybrid search — 30-60% better on complex queries |\n| 8c | [**DiskANN \u002F Vamana**](.\u002Fcrates\u002Fruvector-core) | SSD-backed billion-scale ANN with alpha-RNG pruning and LRU page cache — \u003C10ms latency |\n| 8d | [**ColBERT multi-vector**](.\u002Fcrates\u002Fruvector-core) | Per-token late interaction retrieval with MaxSim, AvgSim, SumMax scoring |\n| 8e | [**Matryoshka embeddings**](.\u002Fcrates\u002Fruvector-core) | Adaptive-dimension search with funnel and cascade modes — speed with minimal recall loss |\n| 8f | [**OPQ**](.\u002Fcrates\u002Fruvector-core) | Optimized Product Quantization with learned rotation — 10-30% error reduction vs standard PQ |\n| 8g | [**LSM compaction**](.\u002Fcrates\u002Fruvector-core) | Log-Structured Merge-tree for write-heavy vector workloads with bloom filters |\n| 8h | [**GraphMAE**](.\u002Fcrates\u002Fruvector-gnn) | Graph Masked Autoencoder — self-supervised node representation learning with GAT encoder |\n| 8i | [**TurboQuant**](.\u002Fcrates\u002Fruvllm) | 2-4 bit asymmetric KV-cache quantization — 6-8x memory reduction, \u003C0.5% perplexity loss, H2O\u002FPyramidKV eviction |\n\n**Continuous Training & Optimization** *(ADR-129)*\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 8j | [**Nightly training**](.\u002Fscripts\u002Ftraining\u002F) | Automated nightly LoRA fine-tuning from brain learnings — models improve every day |\n| 8k | [**Release gates**](.\u002Fscripts\u002Ftraining\u002Frelease_gate.py) | 7 automated quality checks (code quality, routing accuracy, perplexity, speed, contamination) — prevents shipping regressions |\n| 8l | [**TurboQuant profiling**](.\u002Fcrates\u002Fruvllm\u002Fsrc\u002Fquantize\u002Fturboquant_profile.rs) | Per-layer KV-cache bit-width optimization with `.turboquant.json` sidecar configs |\n| 8m | [**Training corpus**](.\u002Fdata\u002Ftraining\u002F) | 230+ records from brain memories (pi.ruv.io) + architecture decisions + Claude routing examples |\n\n**Distributed Systems**\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 9 | [**Raft consensus**](.\u002Fcrates\u002Fruvector-raft) | Leader election and log replication — nodes agree on state even when some fail |\n| 10 | [**Multi-master replication**](.\u002Fcrates\u002Fruvector-replication) | Vector clocks, conflict resolution, geo-distributed sync across data centers |\n| 11 | [**Cluster management**](.\u002Fcrates\u002Fruvector-cluster) | Horizontal scaling with consistent hashing — add nodes without rebalancing everything |\n| 12 | [**Delta consensus**](.\u002Fcrates\u002Fruvector-delta-consensus) | Track behavioral changes across distributed nodes with CRDTs and causal ordering |\n| 13 | **Burst scaling** | 10-50x capacity scaling for traffic spikes — absorb load then scale back down |\n| 14 | **Auto-sharding** | Automatic data partitioning across nodes based on access patterns |\n\n**AI & Machine Learning**\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 15 | [**Run LLMs locally**](.\u002Fcrates\u002Fruvllm) | Load GGUF models and run inference on your hardware — Metal, CUDA, ANE, WebGPU |\n| 16 | [**RuvLTRA models**](https:\u002F\u002Fhuggingface.co\u002Fruv\u002Fruvltra) | Pre-trained GGUF for routing and embeddings in under 10 ms |\n| 17 | [**SONA learning**](.\u002Fcrates\u002Fsona) | Self-Optimizing Neural Architecture — LoRA fine-tuning + EWC++ memory preservation |\n| 18 | [**50+ attention mechanisms**](.\u002Fcrates\u002Fruvector-attention) | FlashAttention-3, MLA, Mamba SSM, KV-cache compression, speculative decoding, graph, hyperbolic, mincut-gated |\n| 19 | [**Semantic routing**](.\u002Fcrates\u002Fruvector-router-core) | Route AI requests to the right model or handler using FastGRNN neural inference |\n| 20 | [**Sparse inference**](.\u002Fcrates\u002Fruvector-sparse-inference) | PowerInfer-style engine — only activate the neurons you need, 2-10x faster on edge |\n| 21 | [**Tiny Dancer**](.\u002Fcrates\u002Fruvector-tiny-dancer-core) | Production-grade agent routing with FastGRNN — lightweight alternative to full LLM |\n| 22 | [**Domain expansion**](.\u002Fcrates\u002Fruvector-domain-expansion) | Cross-domain transfer learning — new tasks bootstrap from past learning automatically |\n| 23 | [**Advanced math**](.\u002Fcrates\u002Fruvector-math) | Optimal transport, Sinkhorn distances, KL divergence, spectral clustering |\n| 24 | [**Coherence measurement**](.\u002Fcrates\u002Fruvector-coherence) | Measure signal quality and compare attention mechanisms objectively |\n| 25 | [**CNN image embeddings**](.\u002Fcrates\u002Fruvector-cnn) | MobileNet-V3 with SIMD\u002FWinograd\u002FINT8 — \u003C5ms image embeddings, WASM-ready, zero deps |\n\n**Graph Transformers** ([8 verified modules](.\u002Fcrates\u002Fruvector-graph-transformer))\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 25 | [**Proof-gated mutation**](.\u002Fcrates\u002Fruvector-verified) | Every write to graph state requires a formal proof — bugs cannot corrupt data |\n| 26 | **Sublinear attention** | O(n log n) via LSH bucketing, PPR sampling, and spectral sparsification |\n| 27 | **Physics-informed layers** | Hamiltonian dynamics, gauge equivariant message passing — energy conserved by construction |\n| 28 | **Biological layers** | Spiking attention, Hebbian\u002FSTDP learning, dendritic branching |\n| 29 | **Self-organizing layers** | Morphogenetic fields, reaction-diffusion growth — graphs that restructure themselves |\n| 30 | **Verified training** | Training certificates, delta-apply rollback — bad gradient steps auto-reversed |\n| 31 | **Manifold geometry** | Product manifolds S^n x H^m x R^k — work in curved spaces, not just flat |\n| 32 | **Temporal-causal layers** | Causal masking, Granger causality extraction, continuous-time ODE integration |\n| 33 | **Economic layers** | Nash equilibrium attention, Shapley attribution — fair value assignment in multi-agent graphs |\n\n**Cognitive Containers** ([RVF format](.\u002Fcrates\u002Frvf\u002FREADME.md))\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 34 | **Self-boot as a microservice** | A single `.rvf` file contains vectors, models, and a Linux kernel — boots in 125 ms |\n| 35 | **eBPF acceleration** | Hot vectors served in kernel data path via XDP, socket filter, and TC programs |\n| 36 | **5.5 KB WASM runtime** | Same `.rvf` file runs queries in a browser tab with zero backend |\n| 37 | [**COW branching**](.\u002Fcrates\u002Frvf) | Git-like copy-on-write — 1M vectors, 100 edits = ~2.5 MB branch |\n| 38 | [**Witness chains**](.\u002Fcrates\u002Frvf\u002Frvf-crypto) | Tamper-evident hash-linked audit trail records every operation automatically |\n| 39 | [**Post-quantum signatures**](.\u002Fcrates\u002Frvf\u002Frvf-crypto) | ML-DSA-65 and SLH-DSA-128s alongside Ed25519 — future-proof cryptography |\n| 40 | **DNA-style lineage** | Track parent\u002Fchild derivation chains with cryptographic hashes |\n| 41 | **25 segment types** | VEC, INDEX, KERNEL, EBPF, WASM, COW_MAP, WITNESS, CRYPTO, and 17 more |\n\n**PostgreSQL Extension** ([230+ SQL functions](.\u002Fcrates\u002Fruvector-postgres))\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 42 | **Drop-in pgvector replacement** | Same SQL interface but with self-learning search — no app changes needed |\n| 43 | **Sublinear solvers in SQL** | PageRank, conjugate gradient, Laplacian solver — O(log n) to O(sqrt(n)) |\n| 44 | **Math distances in SQL** | Wasserstein, Sinkhorn, KL divergence, spectral clustering — all from SQL |\n| 45 | **Topological data analysis** | Persistent homology, Betti numbers, embedding drift detection |\n| 46 | **SONA learning in SQL** | Micro-LoRA trajectory learning with EWC++ forgetting prevention |\n| 47 | **Extended attention in SQL** | O(n) linear, MoE, hyperbolic, sliding window attention — all callable from SQL |\n\n**Specialized Processing**\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 48 | [**SciPix OCR**](.\u002Fexamples\u002Fscipix) | Extract LaTeX and MathML from scientific documents and PDFs |\n| 49 | [**DAG workflows**](.\u002Fcrates\u002Fruvector-dag) | Self-learning directed acyclic graph execution for multi-step pipelines |\n| 50 | [**Cognitum Gate**](.\u002Fcrates\u002Fcognitum-gate-kernel) | Cognitive AI gateway with TileZero acceleration for fast routing |\n| 51 | [**FPGA transformer**](.\u002Fcrates\u002Fruvector-fpga-transformer) | Hardware-accelerated transformer inference on programmable chips |\n| 52 | [**Quantum coherence**](.\u002Fcrates\u002FruQu) | Error correction via dynamic min-cut optimization for quantum circuits |\n| 53 | [**Sublinear solvers**](.\u002Fcrates\u002Fruvector-solver) | 8 algorithms (Neumann, CG, Forward Push, TRUE, BMSSP) — O(log n) to O(sqrt(n)) |\n| 54 | [**Graph sparsifier**](.\u002Fcrates\u002Fruvector-sparsifier) | Compressed shadow graph preserving spectral properties — O(n log n) edges instead of O(n²) |\n| 55 | [**Mincut-gated transformer**](.\u002Fcrates\u002Fruvector-mincut-gated-transformer) | Dynamic attention that prunes irrelevant connections using graph min-cut |\n| 56 | [**Nervous system**](.\u002Fcrates\u002Fruvector-nervous-system) | 5-layer bio-inspired adaptive system with spiking networks and BTSP learning |\n| 57 | [**Prime Radiant**](.\u002Fcrates\u002Fprime-radiant) | Coherence engine using sheaf Laplacian math for AI safety and hallucination detection |\n\n**Genomics & Health** ([rvDNA](.\u002Fexamples\u002Fdna))\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 57 | **rvDNA genomic analysis** | Variant calling, protein translation, HNSW k-mer search in 12 ms |\n| 58 | **`.rvdna` file format** | AI-native binary with pre-computed vectors, tensors, and embeddings |\n| 59 | **Instant diagnostics** | Sickle cell, cancer mutations, drug dosing — runs on any device |\n| 60 | **Privacy-first WASM** | Browser-based genomics — your DNA data never leaves the device |\n| 61 | **Biomarker engine** | Composite polygenic risk scoring (20 SNPs, 6 gene-gene interactions, 2 us) |\n| 62 | **Streaming biomarkers** | Real-time anomaly detection, CUSUM changepoints, trend analysis (>100k readings\u002Fsec) |\n\n**Platform & Integration**\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 63 | **Run anywhere** | Rust, Node.js, browser (WASM), edge ([rvLite](.\u002Fcrates\u002Frvlite)), HTTP server, bare metal |\n| 64 | [**MCP server**](.\u002Fcrates\u002Fmcp-gate) | Model Context Protocol for AI assistants — Claude, GPT, and other agents can use RuVector as a tool |\n| 65 | **Cloud deployment** | One-click deploy to [Google Cloud Run](.\u002Fexamples\u002Fgoogle-cloud), Kubernetes |\n| 66 | [**iOS App Clip**](.\u002Fexamples\u002Fapp-clip) | Scan a QR code to load an RVF cognitive seed on your phone — under 15 MB |\n| 67 | [**Prometheus metrics**](.\u002Fcrates\u002Fruvector-metrics) | Built-in monitoring — export latency, throughput, and memory stats to Grafana |\n| 68 | **90+ Rust crates + npm packages** | Published on [crates.io](https:\u002F\u002Fcrates.io\u002Fcrates\u002Frvf-runtime) and [npm](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Frvf) |\n\n**Examples & Applications**\n| # | Capability | What It Does |\n|---|------------|--------------|\n| 69 | [**Neural Trader**](.\u002Fexamples\u002Fneural-trader) | Algorithmic trading with Kelly Criterion position sizing and LSTM-Transformer prediction |\n| 70 | [**Spiking Neural Network**](.\u002Fexamples\u002Fmeta-cognition-spiking-neural-network) | Hybrid AI combining spiking networks, SIMD vector ops, and 5 attention types |\n| 71 | [**ReFrag Pipeline**](.\u002Fexamples\u002Frefrag-pipeline) | Compress-Sense-Expand architecture — ~30x RAG latency reduction |\n| 72 | [**Edge Network**](.\u002Fexamples\u002Fedge-net) | Distributed collective AI — share idle compute across devices |\n| 73 | [**Vibecast 7Sense**](.\u002Fexamples\u002Fvibecast-7sense) | Transform bird calls into navigable geometric space using vector search |\n| 74 | [**Ultra-Low Latency Sim**](.\u002Fexamples\u002Fultra-low-latency-sim) | Meta-simulation achieving quadrillion simulations per second on CPU with SIMD |\n| 75 | [**Verified Applications**](.\u002Fexamples\u002Fverified-applications) | 10 exotic proof-carrying apps: weapons filters, legal forensics, medical diagnostics |\n\n\u003C\u002Fdetails>\n\n### Built by rUv, powered by [Cognitum.one](https:\u002F\u002Fcognitum.one)\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Cognitum Hardware — The Agentic Appliance & Chip\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n**Cognitum v0 — The Agentic Appliance**: Run tens of thousands of always-on agents at no incremental cost beyond the box. Learns in proximity to any signal — sensors, networks, machines — at near-zero power (~5 uW\u002Finference, \u003C15W total). Sub-millisecond response, 500x cheaper than cloud AI. No cloud bills, no per-agent fees. Like a nervous system, not a brain.\n\n**Cognitum v1 — The Agentic Chip**: Same architecture on a single 257-core custom chip. Runs on less than 2W — a AA battery. Idle-to-8 GHz burst on demand, 2 TB\u002Fs interconnect, built-in encryption per core.\n\n\u003C\u002Fdetails>\n\n### A Complete Agentic AI Operating System\n\nRuVector isn't a database you add to your stack — it's the entire stack. Self-learning, self-optimizing, and self-deploying. Everything an AI application needs to run, from bare metal hardware up to the application layer, in one package:\n\n\n**Intelligence**\n\n| | Layer | Replaces | What It Does |\n|---|-------|----------|--------------|\n| 🔄 | [**Self-Learning**](.\u002Fcrates\u002Fsona\u002FREADME.md) | Manual retraining, MLOps | SONA adapts in \u003C1 ms — LoRA fine-tuning + EWC++ memory on every request |\n| ⚡ | [**Self-Optimizing**](.\u002Fcrates\u002Fruvector-gnn\u002FREADME.md) | Manual tuning, config files | Auto-tunes routing, ranking, compression, and index parameters |\n| 🎯 | [**Embeddings**](.\u002Fcrates\u002Fruvllm\u002FREADME.md) | OpenAI API, Cohere, static models | Contrastive training, triplet loss, real-time fine-tuning — embeddings improve as you use them |\n| ✅ | [**Verified Training**](.\u002Fcrates\u002Fruvector-verified\u002FREADME.md) | Manual validation | Formal proofs + statistical tests on every training step — gradients only apply if invariants pass |\n\n**Data & Search**\n\n| | Layer | Replaces | What It Does |\n|---|-------|----------|--------------|\n| 🔍 | [**Search**](.\u002Fcrates\u002Fruvector-core\u002FREADME.md) | Pinecone, Weaviate, Qdrant | Self-learning HNSW — GNN improves results from every query |\n| 🗄️ | [**Storage**](.\u002Fcrates\u002Fruvector-core\u002FREADME.md) | Separate database + cache | Vector store, graph DB, key-value cache — unified engine |\n| 🐘 | [**PostgreSQL**](.\u002Fcrates\u002Fruvector-postgres\u002FREADME.md) | pgvector, pg_embedding | Drop-in replacement — 230+ SQL functions, same interface but search gets smarter over time |\n| 🔗 | [**Graph**](.\u002Fcrates\u002Fruvector-graph\u002FREADME.md) | Neo4j, Amazon Neptune | Cypher, W3C SPARQL 1.1, hyperedges — all built in |\n\n**AI & ML**\n\n| | Layer | Replaces | What It Does |\n|---|-------|----------|--------------|\n| 🤖 | [**AI Runtime**](.\u002Fcrates\u002Fruvllm\u002FREADME.md) | llama.cpp, vLLM, Ollama | ruvllm — GGUF models, MicroLoRA (\u003C1 ms), speculative decoding, continuous batching, WASM |\n| 🧠 | [**ML Framework**](.\u002Fcrates\u002Fruvector-attention\u002FREADME.md) | PyTorch, TensorFlow | 50+ attention types, FlashAttention-3, MLA, Mamba SSM, Graph RAG, DiskANN, 8 graph transformers, sublinear solvers |\n| 🔬 | [**Coherence**](.\u002Fcrates\u002Fruvector-mincut\u002FREADME.md) | Manual testing, guardrails | Min-cut finds the weakest links in any network — detects AI drift, prunes wasted compute (50% reduction), keeps agents in sync |\n| 🧬 | [**Domain Models**](.\u002Fcrates\u002Fruvector-domain-expansion\u002FREADME.md) | Custom ML pipelines | Genomics (DNA variant calling), physics simulation, economic modeling, biological networks |\n\n**Infrastructure**\n\n| | Layer | Replaces | What It Does |\n|---|-------|----------|--------------|\n| 🔧 | [**Hardware**](.\u002Fcrates\u002Fruvector-fpga-transformer\u002FREADME.md) | CUDA toolkit, driver configs | Sparse\u002Fspiking CPU (AVX-512, NEON) — GPU for bursts (Metal, CUDA, ANE, WebGPU, FPGA) |\n| 🐧 | [**Kernel**](.\u002Fcrates\u002Frvf\u002FREADME.md) | Linux + Docker + eBPF | `.rvf` file boots its own kernel in 125 ms — eBPF accelerates hot paths |\n| 🌐 | [**Coordination**](.\u002Fcrates\u002Fruvector-raft\u002FREADME.md) | etcd, ZooKeeper, Consul | Raft consensus, multi-master replication, CRDT delta sync, auto-sharding |\n| 📦 | [**Packaging**](.\u002Fcrates\u002Frvf\u002FREADME.md) | Docker, Kubernetes | One `.rvf` file = your entire service — servers, browsers, phones, IoT, bare metal |\n\n**Routing & Observability**\n\n| | Layer | Replaces | What It Does |\n|---|-------|----------|--------------|\n| 🚦 | [**Routing**](.\u002Fcrates\u002Fruvector-tiny-dancer-core\u002FREADME.md) | API gateways, LLM routers | Semantic routing (Tiny Dancer), MCP protocol gateway, agent-to-agent discovery |\n| 📊 | [**Observability**](.\u002Fcrates\u002Fruvector-profiler\u002FREADME.md) | Datadog, Prometheus | Latency\u002Fpower\u002Fmemory profiling, coherence scoring, real-time metrics |\n| 🛡️ | [**Safety**](.\u002Fcrates\u002Fcognitum-gate-tilezero\u002FREADME.md) | Manual review, guardrails | Cognitum Gate — 256-tile WASM fabric, Permit\u002FDefer\u002FDeny in \u003C1 ms, witness receipts |\n\n**Security & Trust**\n\n| | Layer | Replaces | What It Does |\n|---|-------|----------|--------------|\n| 🔐 | [**Crypto**](.\u002Fcrates\u002Frvf\u002Frvf-crypto\u002FREADME.md) | Vault, manual audit logs | Post-quantum (ML-DSA-65, Ed25519), SHAKE-256, witness chains, hardware attestation |\n| 📜 | [**Lineage**](.\u002Fcrates\u002Frvf\u002Frvf-crypto\u002FREADME.md) | No equivalent | Every operation recorded in a tamper-proof chain — full provenance from creation to deployment |\n\nThe [RVF cognitive container](.\u002Fcrates\u002Frvf\u002FREADME.md) ties it all together: a single file that packages your vectors, models, data, and a bootable kernel. Drop it on any machine and it starts serving in 125 ms — no install, no dependencies. It branches like Git (only changes are copied), logs every operation in a tamper-proof chain, and runs anywhere from a browser to bare metal.\n\n---\n\n## How the GNN Works\n\nTraditional vector search:\n```\nQuery → HNSW Index → Top K Results\n```\n\nRuVector with GNN:\n```\nQuery → HNSW Index → GNN Layer → Enhanced Results\n                ↑                      │\n                └──── learns from ─────┘\n```\n\nThe GNN layer:\n1. Takes your query and its nearest neighbors\n2. Applies multi-head attention to weigh which neighbors matter\n3. Updates representations based on graph structure\n4. Returns better-ranked results — all in under 1ms\n\nThis is **temporal learning** — the system learns from the sequence and timing of queries, not just their content. A query asked right after another carries context. Patterns that repeat get reinforced. Paths that lead to good results get stronger over time. The result: search gets faster and more accurate the more you use it, adapting in real time without retraining.\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Deep Dive: How Self-Learning Search Actually Works\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n### The Problem with Normal Search\n\nEvery vector database does the same thing: you give it a query, it finds the closest matches by distance, and returns them. The results never change. Search the same thing a thousand times and you get the same answer a thousand times — even if the first result was wrong and you always clicked the third one instead.\n\nRuVector is different. It watches what happens *after* the search and uses that to make the next search better.\n\n### What the GNN Actually Does\n\nThink of your data as a city map. Each vector is a building, and the HNSW index creates roads between similar buildings. A normal search just walks the shortest road to find nearby buildings.\n\nThe GNN is like a local who knows the shortcuts. It looks at the neighborhood around your destination and says: \"Yes, that building is close, but *this* one over here is what you actually want.\" It learns these shortcuts by watching which results people actually use.\n\n**Technically, it works in three steps:**\n\n| Step | What Happens | Plain English |\n|------|-------------|---------------|\n| **1. Message Passing** | Each node collects information from its HNSW neighbors | \"Ask the neighborhood what they know\" |\n| **2. Attention Weighting** | Multi-head attention scores which neighbors matter most for this specific query | \"Some neighbors are more helpful than others — figure out which ones\" |\n| **3. Representation Update** | Node representations shift based on what the neighborhood says | \"Update your understanding based on what you learned\" |\n\nThis entire process takes **under 1ms** thanks to SIMD acceleration (processing 4-8 numbers at once instead of one at a time).\n\n### Temporal Learning: Time Matters\n\nMost AI systems treat every input as independent — they don't know or care what happened 5 seconds ago. RuVector tracks the *sequence* and *timing* of queries, which reveals patterns that individual queries can't:\n\n| Pattern | What It Reveals | How RuVector Adapts |\n|---------|----------------|---------------------|\n| Same user searches A then B within seconds | A and B are related, even if they're far apart in vector space | Strengthens the path between A and B |\n| Many users skip result #1 and click result #3 | Result #3 is actually more relevant | GNN learns to rank #3 higher next time |\n| Query bursts around a topic at certain times | Temporal relevance — some things matter more at certain times | Boosts recently-active paths |\n| A query that follows a specific sequence | Context from previous queries changes what \"good results\" means | Attention weights shift based on session context |\n\n### Three Types of Learning\n\nRuVector learns at three different speeds simultaneously:\n\n| Speed | Mechanism | What It Does | Latency |\n|-------|-----------|-------------|---------|\n| **Instant** | MicroLoRA adaptation | Adjusts weights for this specific request based on immediate feedback | \u003C1ms |\n| **Session** | GNN attention updates | Reinforces paths that led to good results during this session | ~10ms (background) |\n| **Long-term** | EWC++ consolidation | Permanently strengthens important patterns without forgetting old ones | ~100ms (background) |\n\nThe key innovation is **EWC++ (Elastic Weight Consolidation)** — it solves the \"catastrophic forgetting\" problem. Without it, learning new patterns would erase old ones. EWC++ identifies which weights are important for existing knowledge and protects them while still allowing new learning.\n\n### Why It's Fast: The HNSW Shortcut\n\nThe GNN doesn't run on your entire dataset. It only runs on the small subgraph of HNSW neighbors that are relevant to the current query — typically 10-50 nodes out of millions. This is why it adds under 1ms of latency instead of seconds:\n\n```\n1M vectors in your database\n    → HNSW finds ~50 candidate neighbors        (0.3ms)\n    → GNN re-ranks those 50 with attention       (0.4ms)\n    → Return top K results                       (0.1ms)\n    ──────────────────────────────────────────\n    Total: \u003C1ms, and results improve over time\n```\n\n### What Improves Over Time\n\n| Metric | Day 1 | After 1K Queries | After 100K Queries |\n|--------|-------|------------------|-------------------|\n| **Recall@10** | Baseline (HNSW only) | +5-8% | +12.4% |\n| **Query latency** | ~0.8ms | ~0.7ms (hot paths cached) | ~0.5ms (optimized routing) |\n| **Relevance** | Distance-based only | Learns user preferences | Personalized per query pattern |\n\n### Three GNN Architectures (Pick One or Stack Them)\n\n| Architecture | Best For | How It Works |\n|-------------|----------|-------------|\n| **GCN** (Graph Convolutional Network) | General-purpose re-ranking | Averages neighbor information — simple, fast, effective |\n| **GAT** (Graph Attention Network) | Queries where some neighbors matter more than others | Learns *which* neighbors to pay attention to per query |\n| **GraphSAGE** | Datasets that change frequently (new vectors added often) | Can score new vectors it's never seen before, without retraining |\n\n### Runs Everywhere\n\nThe same GNN code runs natively in Rust, in Node.js via NAPI-RS bindings, and in the browser via WebAssembly. Models trained on the server can be exported and run client-side — a user's browser can do personalized re-ranking without sending queries to a server.\n\n\u003C\u002Fdetails>\n\n## Quick Start\n\n### One-Line Install\n\n```bash\n# Interactive installer - lists all packages\nnpx ruvector install\n\n# Or install directly\nnpm install ruvector\nnpx ruvector\n\n# Self-learning hooks for Claude Code\nnpx @ruvector\u002Fcli hooks init\nnpx @ruvector\u002Fcli hooks install\n\n# LLM runtime (SONA learning, HNSW memory)\nnpm install @ruvector\u002Fruvllm\n```\n\n### Node.js \u002F Browser\n\n```bash\n# Install\nnpm install ruvector\n\n# Or try instantly\nnpx ruvector\n```\n\n---\n\n### Ecosystem: AI Agent Orchestration\n\nRuVector powers two major AI orchestration platforms:\n\n| Platform | Purpose | Install | Downloads |\n|----------|---------|---------|-----------|\n| [**ruFlo**](https:\u002F\u002Fgithub.com\u002Fruvnet\u002Fclaude-flow) | Enterprise multi-agent orchestration for Claude Code | `npx ruvflo@latest` | [![npm downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdt\u002Fclaude-flow.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fclaude-flow) |\n| [**Agentic-Flow**](https:\u002F\u002Fgithub.com\u002Fruvnet\u002Fagentic-flow) | Run AI agents on any cloud with any model — Claude, GPT, Gemini, or local | `npx agentic-flow@latest` | [![npm downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdt\u002Fagentic-flow.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fagentic-flow) |\n| [**AgentDB**](https:\u002F\u002Fgithub.com\u002Fruvnet\u002Fagentdb) | Give AI agents long-term memory that gets smarter over time | `npm install agentdb@alpha` | [![npm downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdt\u002Fagentdb.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fagentdb) |\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Claude-Flow v3\u003C\u002Fstrong> — Turn Claude Code into a collaborative AI team\u003C\u002Fsummary>\n\n**54+ specialized agents** working together on complex software engineering tasks:\n\n```bash\n# Install\nnpx ruvflo@latest init --wizard\n\n# Spawn a swarm\nnpx ruvflo@latest swarm init --topology hierarchical --max-agents 8\n```\n\n**Key Features:**\n- **SONA Learning**: Sub-50ms adaptive routing, learns optimal patterns over time\n- **Queen-led Swarms**: Byzantine fault-tolerant consensus with 5 protocols (Raft, Gossip, CRDT)\n- **HNSW Memory**: 150x-12,500x faster pattern retrieval via RuVector\n- **175+ MCP Tools**: Native Model Context Protocol integration\n- **Cost Optimization**: 3-tier routing extends Claude Code quota by 2.5x\n- **Security**: AIDefence threat detection (\u003C10ms), prompt injection blocking\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Agentic-Flow v2\u003C\u002Fstrong> — Production AI agents for any cloud\u003C\u002Fsummary>\n\n**66 self-learning agents** with Claude Agent SDK, deployable to any cloud:\n\n```bash\n# Install\nnpx agentic-flow@latest\n\n# Or with npm\nnpm install agentic-flow\n```\n\n**Key Features:**\n- **SONA Architecture**: \u003C1ms adaptive learning, +55% quality improvement\n- **Flash Attention**: 2.49x JS speedup, 7.47x with NAPI bindings\n- **213 MCP Tools**: Swarm management, memory, GitHub integration\n- **Agent Booster**: 352x faster code editing for simple transforms\n- **Multi-Provider**: Claude, GPT, Gemini, Cohere, local models with failover\n- **Graph Reasoning**: GNN query refinement with +12.4% recall improvement\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>rvDNA\u003C\u002Fstrong> — AI-native genomic diagnostics, instant and available to everyone\u003C\u002Fsummary>\n\n**Using AI to make the world a healthier place.** rvDNA puts genomic diagnostics on any device — a phone, a laptop, a browser tab — in 12 milliseconds. No cloud, no GPU, no subscription. Private by default.\n\n```bash\ncargo add rvdna              # Rust\nnpm install @ruvector\u002Frvdna  # JavaScript \u002F TypeScript\n```\n\n| What It Does | How |\n|---|---|\n| Find mutations (sickle cell, cancer) | Bayesian variant calling, 155 ns\u002FSNP |\n| Translate DNA to protein | Full codon table + GNN contact graphs |\n| Predict biological age | Horvath clock, 353 CpG sites |\n| Recommend drug doses | CYP2D6 star alleles + CPIC guidelines |\n| Score health risks | 20 SNPs, 6 gene-gene interactions, composite risk scoring in 2 us |\n| Stream biomarker data | Real-time anomaly detection, CUSUM changepoints, >100k readings\u002Fsec |\n| Search genomes by similarity | HNSW k-mer vectors, O(log N) |\n| Store pre-computed AI features | `.rvdna` binary format — open and instant |\n\n- **Rust crate**: [crates.io\u002Fcrates\u002Frvdna](https:\u002F\u002Fcrates.io\u002Fcrates\u002Frvdna)\n- **npm package**: [@ruvector\u002Frvdna](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Frvdna) (NAPI-RS native + JS fallback)\n- **Source**: [examples\u002Fdna](.\u002Fexamples\u002Fdna)\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>RVF Cognitive Containers\u003C\u002Fstrong> — One file that stores, boots, and proves everything\u003C\u002Fsummary>\n\n**[RVF (RuVector Format)](.\u002Fcrates\u002Frvf\u002FREADME.md)** is a universal binary substrate that merges database, model, graph engine, kernel, and attestation into a single deployable file. A `.rvf` file can store vector embeddings, carry LoRA adapter deltas, embed GNN graph state, include a bootable Linux microkernel, run queries in a 5.5 KB WASM runtime, and prove every operation through a cryptographic witness chain — all in one file that runs anywhere from a browser to bare metal.\n\nThis is not a database format. It is an **executable knowledge unit**.\n\n```bash\ncargo install rvf-cli                          # CLI tool\ncargo add rvf-runtime                          # Rust library\nnpm install @ruvector\u002Frvf                      # TypeScript SDK\nnpx @ruvector\u002Frvf-mcp-server --transport stdio # MCP server for AI agents\n```\n\n| What It Does | How |\n|---|---|\n| Self-boot as a microservice | Real Linux kernel in the file, boots in 125 ms on QEMU\u002FKVM |\n| Hardware-speed lookups | eBPF programs (XDP, TC, socket filter) bypass userspace entirely |\n| Run in any browser | 5.5 KB WASM runtime, zero backend |\n| Git-like branching | COW at cluster granularity — 1M vectors, 100 edits = ~2.5 MB child |\n| Tamper-evident audit | Hash-linked witness chain for every insert, query, and deletion |\n| Post-quantum signatures | ML-DSA-65 and Ed25519 signing on every segment |\n| DNA-style lineage | Parent\u002Fchild derivation chains with cryptographic verification |\n| 28 segment types | VEC, INDEX, KERNEL, EBPF, WASM, COW_MAP, WITNESS, CRYPTO, FEDERATED_MANIFEST, and 19 more |\n\n**Rust crates** (23): [`rvf-types`](https:\u002F\u002Fcrates.io\u002Fcrates\u002Frvf-types) `rvf-wire` `rvf-manifest` `rvf-quant` `rvf-index` `rvf-crypto` [`rvf-runtime`](https:\u002F\u002Fcrates.io\u002Fcrates\u002Frvf-runtime) `rvf-kernel` `rvf-ebpf` [`rvf-federation`](.\u002Fcrates\u002Frvf\u002Frvf-federation) `rvf-launch` `rvf-server` `rvf-import` [`rvf-cli`](https:\u002F\u002Fcrates.io\u002Fcrates\u002Frvf-cli) `rvf-wasm` `rvf-solver-wasm` `rvf-node` + 6 adapters (claude-flow, agentdb, ospipe, agentic-flow, rvlite, sona)\n\n**npm packages** (6): [`@ruvector\u002Frvf`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Frvf) [`@ruvector\u002Frvf-node`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Frvf-node) [`@ruvector\u002Frvf-wasm`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Frvf-wasm) [`@ruvector\u002Frvf-mcp-server`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Frvf-mcp-server) [`@ruvector\u002Fruvllm-wasm`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Fruvllm-wasm) [`@ruvector\u002Fneural-trader-wasm`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Fneural-trader-wasm)\n\n- **Security Hardened RVF** ([`examples\u002Fsecurity_hardened.rvf`](.\u002Fexamples\u002Fsecurity_hardened.rvf)) — 2.1 MB sealed artifact with 22 verified capabilities: TEE attestation (SGX\u002FSEV-SNP\u002FTDX\u002FARM CCA), AIDefence (injection\u002Fjailbreak\u002FPII\u002Fexfil), hardened Linux microkernel, eBPF firewall, Ed25519 signing, 6-role RBAC, Coherence Gate, 30-entry witness chain, Paranoid policy, COW branching, audited k-NN. See [ADR-042](.\u002Fdocs\u002Fadr\u002FADR-042-Security-RVF-AIDefence-TEE.md).\n- **Full documentation**: [crates\u002Frvf\u002FREADME.md](.\u002Fcrates\u002Frvf\u002FREADME.md)\n- **ADR-030**: [Cognitive Container Architecture](.\u002Fdocs\u002Fadr\u002FADR-030-rvf-cognitive-container.md)\n- **ADR-031**: [COW Branching & Real Containers](.\u002Fdocs\u002Fadr\u002FADR-031-rvcow-branching-and-real-cognitive-containers.md)\n- **ADR-042**: [Security RVF — AIDefence + TEE](.\u002Fdocs\u002Fadr\u002FADR-042-Security-RVF-AIDefence-TEE.md)\n- **56 runnable examples**: [examples\u002Frvf\u002Fexamples\u002F](.\u002Fexamples\u002Frvf\u002Fexamples\u002F)\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>RuVix Cognition Kernel\u003C\u002Fstrong> — an operating system for AI agents\u003C\u002Fsummary>\n\n**[RuVix](.\u002Fcrates\u002Fruvix\u002FREADME.md)** is a cognition kernel designed for agentic workloads. Where Linux thinks in files, processes, and POSIX syscalls, RuVix thinks in vectors, graphs, proofs, and capabilities. Every mutation is proof-gated — no cryptographic proof, no state change. Every resource access goes through unforgeable capability tokens. The result: a kernel that understands AI workloads natively.\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│                    AGENT CONTROL PLANE                          │\n│  Claude Code │ AgentDB │ Custom Agents │ RuVLLM Inference       │\n├─────────────────────────────────────────────────────────────────┤\n│                    RVF COMPONENT SPACE                          │\n│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐           │\n│  │ RuView   │ │ AgentDB  │ │ RuVLLM   │ │ Network  │ ...       │\n│  │ Percept. │ │ Intelli. │ │ Infer.   │ │ Stack    │           │\n│  └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘           │\n│       │queue       │queue       │queue       │queue             │\n├───────┴────────────┴────────────┴────────────┴──────────────────┤\n│                    RUVIX COGNITION KERNEL                       │\n│  Capability Manager │ Queue IPC │ Coherence-Aware Scheduler     │\n│  Region Memory │ Proof Engine │ Vector\u002FGraph Kernel Objects     │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n| Primitive | Purpose | Status |\n|-----------|---------|--------|\n| **Task** | Concurrent execution with capability set | ✅ Implemented |\n| **Capability** | Unforgeable access token (seL4-inspired) | ✅ Implemented |\n| **Region** | Memory with policy (immutable\u002Fappend\u002Fslab) | ✅ Implemented |\n| **Queue** | io_uring-style lock-free IPC | ✅ Implemented |\n| **Timer** | Deadline-driven scheduling | ✅ Implemented |\n| **Proof** | Cryptographic attestation for mutations | ✅ Implemented |\n\n**Phases A-F complete**: 20+ crates, 1,200+ tests, 12 syscalls, full deterministic replay.\n\n| Phase | Components | Description |\n|-------|------------|-------------|\n| **A** | Core Kernel | 9 crates: types, cap, region, queue, proof, sched, boot, vecgraph, nucleus |\n| **B** | Bare Metal | AArch64 boot, MMU, exception vectors, HAL, physical memory allocator |\n| **C** | Multi-Core | SMP support (256 cores), ticket spinlocks, IPIs, DMA, Device Tree |\n| **D** | Raspberry Pi | BCM2711\u002F2712 drivers, GPIO, VideoCore mailbox, config.txt parsing |\n| **E** | Network\u002FFS | Ethernet\u002FIPv4\u002FUDP stack, VFS layer, FAT32, RamFS |\n| **F** | Tooling | CLI, kernel shell, QEMU swarm simulation with PBFT consensus |\n\n**Developer Tools**:\n\n| Tool | Purpose |\n|------|---------|\n| `ruvix build` | Cross-compile kernel for AArch64 targets |\n| `ruvix flash` | Flash kernel to SD card or network boot |\n| `ruvix keys` | Ed25519 key management for secure boot |\n| `ruvix dtb` | Device tree validation, comparison, decompilation |\n| `ruvix monitor` | Real-time kernel metrics dashboard |\n| `ruvix security` | Security audit and CVE scanning |\n| `rvsh` (kernel) | In-kernel debug shell with 13 commands |\n\n**Distributed Testing (QEMU Swarm)**:\n\n| Feature | Description |\n|---------|-------------|\n| Multi-node clusters | Spawn N QEMU instances as cluster nodes |\n| PBFT consensus | Byzantine fault-tolerant consensus (f \u003C n\u002F3) |\n| Virtual networking | Mesh, ring, star topologies |\n| Fault injection | Network partitions, node crashes, latency |\n| Console multiplexing | Aggregate output from all nodes |\n\n- **Full documentation**: [crates\u002Fruvix\u002FREADME.md](.\u002Fcrates\u002Fruvix\u002FREADME.md)\n- **ADR-087**: [RuVix Cognition Kernel Architecture](.\u002Fdocs\u002Fadr\u002FADR-087-ruvix-cognition-kernel.md)\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Sublinear-Time Solver\u003C\u002Fstrong> — math that gets faster as your data gets bigger\u003C\u002Fsummary>\n\n**[ruvector-solver](.\u002Fcrates\u002Fruvector-solver\u002FREADME.md)** solves large math problems (like ranking pages, finding connections in graphs, or computing AI attention) in a fraction of the time traditional solvers need. Where standard approaches slow down dramatically with scale (doubling data = 8x slower), RuVector's 8 specialized algorithms barely notice the increase (doubling data = barely any slower). This is what powers the self-learning engine — fast graph math is what lets search improve in real time instead of waiting minutes to retrain.\n\n```bash\ncargo add ruvector-solver --features all-algorithms\n```\n\n| Algorithm | Complexity | Best For |\n|-----------|-----------|----------|\n| **Neumann Series** | O(k · nnz) | Diagonally dominant, fast convergence |\n| **Conjugate Gradient** | O(√κ · log(1\u002Fε) · nnz) | Gold-standard SPD solver |\n| **Forward Push** | O(1\u002Fε) | Single-source PageRank |\n| **Backward Push** | O(1\u002Fε) | Reverse relevance computation |\n| **Hybrid Random Walk** | O(√n\u002Fε) | Pairwise relevance, Monte Carlo |\n| **TRUE** | O(log n) amortized | Large-scale Laplacian systems |\n| **BMSSP** | O(nnz · log n) | Multigrid hierarchical solve |\n| **Auto Router** | Automatic | Selects optimal algorithm |\n\n**Key optimizations**: AVX2 SIMD SpMV, fused residual kernels, bounds-check elimination, arena allocator\n\n**Supporting crates**:\n- [`ruvector-attn-mincut`](.\u002Fcrates\u002Fruvector-attn-mincut\u002FREADME.md) — Min-cut gating as alternative to softmax attention\n- [`ruvector-coherence`](.\u002Fcrates\u002Fruvector-coherence\u002FREADME.md) — Coherence measurement for attention comparison\n- [`ruvector-profiler`](.\u002Fcrates\u002Fruvector-profiler\u002FREADME.md) — Memory, power, and latency benchmarking\n\n- **177 tests** | 5 Criterion benchmarks | WASM + NAPI bindings\n- **ADR documentation**: [docs\u002Fresearch\u002Fsublinear-time-solver\u002F](.\u002Fdocs\u002Fresearch\u002Fsublinear-time-solver\u002F)\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Graph Sparsifier\u003C\u002Fstrong> — a small graph that behaves like the big one\u003C\u002Fsummary>\n\n**[ruvector-sparsifier](.\u002Fcrates\u002Fruvector-sparsifier\u002FREADME.md)** maintains a compressed \"shadow graph\" that preserves the structural properties of a much larger graph. Think of it as a lossy summary: instead of storing every connection between millions of nodes, the sparsifier keeps only the most important edges — enough to answer questions about cuts, connectivity, and flow almost as accurately as the full graph, but with far less memory and compute.\n\nIt updates incrementally as edges are added or removed, so the compressed version stays current without rebuilding from scratch. A built-in auditor periodically checks that the approximation is still good; if quality drifts, it triggers an automatic rebuild.\n\n```bash\ncargo add ruvector-sparsifier\n```\n\n| Component | What it does |\n|-----------|-------------|\n| **Backbone** | Spanning forest (union-find) that guarantees connectivity is never lost |\n| **Importance scorer** | Random walks estimate which edges matter most for the graph's structure |\n| **Spectral sampler** | Keeps edges proportional to their importance, reweights to stay unbiased |\n| **Auditor** | Random probes verify the compressed graph still matches the original |\n\n- **49 tests** | Criterion benchmarks | WASM bindings\n- **Research**: [docs\u002Fresearch\u002Fspectral-sparsification\u002F](.\u002Fdocs\u002Fresearch\u002Fspectral-sparsification\u002F)\n\n\u003C\u002Fdetails>\n\n---\n\n## How RuVector Compares\n\nSee how RuVector stacks up against popular vector databases across 40+ features — from latency and graph queries to self-learning, cognitive containers, and PostgreSQL integration.\n\n\u003Cdetails>\n\u003Csummary>📊 Comparison with Other Vector Databases\u003C\u002Fsummary>\n\nGrouped comparison across 10 categories. RuVector is the only vector database that learns from usage, runs AI locally, and ships as a single self-booting file.\n\n**Performance & Storage**\n| Feature | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---------|----------|----------|--------|--------|----------|----------|\n| Latency (p50) | **61 us** | ~2 ms | ~1 ms | ~5 ms | ~50 ms | ~5 ms |\n| Memory (1M vectors) | **200 MB*** | 2 GB | 1.5 GB | 1 GB | 3 GB | 1.5 GB |\n| SIMD acceleration | AVX-512, NEON | Partial | ✅ | ✅ | ❌ | Partial |\n| Auto-compression | 2-32x adaptive | ❌ | ❌ | ✅ | ❌ | PQ only |\n| Temporal tensor compression | 4-10x reuse | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Sparse vectors (BM25\u002FTF-IDF) | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ |\n\n**Search & Query**\n| Feature | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---------|----------|----------|--------|--------|----------|----------|\n| Vector similarity search | ✅ HNSW | ✅ | ✅ HNSW | ✅ HNSW | ✅ | ✅ HNSW |\n| Metadata filtering | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |\n| Graph queries (Cypher) | ✅ full engine | ❌ | ❌ | ❌ | ❌ | ❌ |\n| SPARQL\u002FRDF (W3C 1.1) | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Hyperedges (3+ node) | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Hyperbolic embeddings | Poincare + Lorentz | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Multi-tenancy | ✅ collections | ✅ namespaces | ✅ | ✅ | ✅ | ✅ |\n\n**Self-Learning & AI** — features unique to RuVector\n| Feature | RuVector | All Others |\n|---------|----------|------------|\n| GNN on HNSW — search improves from usage | ✅ every query teaches the index | ❌ static index |\n| SONA runtime adaptation | ✅ LoRA + EWC++ auto-tuning | ❌ manual tuning |\n| 46 attention mechanisms | Flash, linear, graph, hyperbolic, mincut-gated | ❌ |\n| Semantic routing (Tiny Dancer) | FastGRNN neural agent routing | ❌ |\n| Sparse inference (PowerInfer-style) | 2-10x faster on edge devices | ❌ |\n| Domain expansion | Cross-domain transfer learning with bandits | ❌ |\n| Self-learning hooks | Q-learning, neural patterns, HNSW memory | ❌ |\n| ReasoningBank | Trajectory learning with verdict judgment | ❌ |\n\n**Local AI — no cloud APIs needed**\n| Feature | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---------|----------|----------|--------|--------|----------|----------|\n| Built-in LLM runtime | ✅ ruvllm (GGUF) | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Hardware acceleration | Metal, CUDA, ANE, WebGPU | N\u002FA | N\u002FA | GPU indexing | N\u002FA | N\u002FA |\n| Pre-trained models | [RuvLTRA](https:\u002F\u002Fhuggingface.co\u002Fruv\u002Fruvltra) (\u003C10 ms) | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Local ONNX embeddings | 8+ models, no API calls | ❌ | ❌ | ❌ | ❌ | text2vec modules |\n| MCP server for AI agents | ✅ mcp-gate | ❌ | ❌ | ❌ | ❌ | ❌ |\n\n**Graph Transformers** — verified graph neural network modules\n| Feature | RuVector | All Others |\n|---------|----------|------------|\n| Proof-gated mutation | Every write requires a formal proof — bugs cannot corrupt | ❌ |\n| Sublinear attention | O(n log n) via LSH, PPR, spectral sparsification | ❌ |\n| Physics-informed layers | Hamiltonian dynamics, energy conserved by construction | ❌ |\n| Biological layers | Spiking, Hebbian\u002FSTDP, dendritic branching | ❌ |\n| Manifold geometry | Product manifolds S^n x H^m x R^k | ❌ |\n| Temporal-causal layers | Granger causality, continuous-time ODE | ❌ |\n| Economic layers | Nash equilibrium, Shapley attribution | ❌ |\n| Verified training | Certificates, delta-apply rollback, fail-closed | ❌ |\n\n**Math & Solvers**\n| Feature | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---------|----------|----------|--------|--------|----------|----------|\n| Sublinear solvers (8 algorithms) | O(log n) to O(sqrt(n)) | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Dynamic min-cut | n^0.12 complexity | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Optimal transport distances | Wasserstein, Sinkhorn, KL | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Topological data analysis | Persistent homology, Betti numbers | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Coherence measurement | Prime Radiant sheaf Laplacian | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Quantum error correction | ruQu dynamic min-cut | ❌ | ❌ | ❌ | ❌ | ❌ |\n\n**Distributed Systems**\n| Feature | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---------|----------|----------|--------|--------|----------|----------|\n| Raft consensus | ✅ | ❌ managed | ✅ | ❌ | ❌ | ✅ |\n| Multi-master replication | ✅ vector clocks | ❌ | ❌ | ✅ | ❌ | ✅ |\n| Auto-sharding | ✅ consistent hashing | ✅ managed | ✅ | ✅ | ❌ | ✅ |\n| Delta consensus (CRDT) | ✅ causal ordering | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Burst scaling (10-50x) | ✅ | ✅ managed | ❌ | ✅ | ❌ | ❌ |\n| Snapshots \u002F backups | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ |\n| Streaming API | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ |\n\n**Cognitive Containers (RVF)** — single-file deployment unique to RuVector\n| Feature | RuVector | All Others |\n|---------|----------|------------|\n| Self-booting microservice | `.rvf` file boots in 125 ms with Linux kernel | ❌ requires server setup |\n| eBPF acceleration | XDP, socket filter, TC kernel data path | ❌ |\n| COW branching | Git-like — 1M vectors, 100 edits = ~2.5 MB branch | ❌ copy everything |\n| Witness chains | Tamper-evident hash-linked audit trail | ❌ manual logging |\n| Post-quantum signatures | ML-DSA-65, SLH-DSA-128s, Ed25519 | ❌ |\n| 25 segment types | VEC, INDEX, KERNEL, EBPF, WASM, COW_MAP, and 19 more | ❌ |\n\n**Platform & Deployment**\n| Feature | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---------|----------|----------|--------|--------|----------|----------|\n| Browser \u002F WASM | ✅ WebGPU, 58 KB | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Edge standalone | ✅ rvLite | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Node.js native | ✅ NAPI-RS | ❌ | Client only | Client only | ✅ | Client only |\n| PostgreSQL extension | ✅ 230+ SQL functions | ❌ | ❌ | ❌ | ❌ | ❌ |\n| iOS App Clip | ✅ QR → RVF in \u003C15 MB | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Cloud deployment | Cloud Run, Kubernetes | Managed only | Docker, K8s | Docker, K8s | Docker | Docker, K8s |\n| FPGA acceleration | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |\n| Prometheus metrics | ✅ built-in | Dashboard | ✅ | ✅ | ❌ | ✅ |\n\n**Specialized Applications**\n| Feature | RuVector | All Others |\n|---------|----------|------------|\n| Genomics (rvDNA) | Variant calling, k-mer search in 12 ms, browser WASM | ❌ |\n| Neural trading | Kelly Criterion + LSTM-Transformer prediction | ❌ |\n| Scientific OCR (SciPix) | LaTeX\u002FMathML extraction from papers | ❌ |\n| Spiking neural networks | Neuromorphic computing, BTSP learning | ❌ |\n| Bio-inspired nervous system | 5-layer adaptive system with EWC plasticity | ❌ |\n| DAG workflows | Self-learning directed graph execution | ❌ |\n| Cognitum Gate | Cognitive AI gateway with TileZero acceleration | ❌ |\n\n**Licensing & Cost**\n| | RuVector | Pinecone | Qdrant | Milvus | ChromaDB | Weaviate |\n|---|----------|----------|--------|--------|----------|----------|\n| License | MIT (free forever) | Proprietary | Apache 2.0 | Apache 2.0 | Apache 2.0 | BSD-3 |\n| Self-hosted | ✅ | ❌ managed only | ✅ | ✅ | ✅ | ✅ |\n| Pricing model | Free | Per-vector\u002Fquery | Free + Cloud | Free + managed | Free + Cloud | Free + Cloud |\n\n\\* Memory with PQ8 compression. Benchmarks on Apple M2 \u002F Intel i7.\n\n\u003C\u002Fdetails>\n\n## Features\n\nEverything RuVector can do — organized by category. Vector search, graph queries, LLM inference, distributed systems, deployment targets, and the self-learning stack that ties it all together.\n\n\u003Cdetails>\n\u003Csummary>⚡ Core Features & Capabilities\u003C\u002Fsummary>\n\n### Core Capabilities\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **Vector Search** | HNSW index, \u003C0.5ms latency, SIMD acceleration | Fast enough for real-time apps |\n| **Cypher Queries** | `MATCH`, `WHERE`, `CREATE`, `RETURN` | Familiar Neo4j syntax |\n| **GNN Layers** | Neural network on index topology | Search improves with usage |\n| **Hyperedges** | Connect 3+ nodes at once | Model complex relationships |\n| **Metadata Filtering** | Filter vectors by properties | Combine semantic + structured search |\n| **Collections** | Namespace isolation, multi-tenancy | Organize vectors by project\u002Fuser |\n| **Hyperbolic HNSW** | Poincaré ball indexing for hierarchies | Better tree\u002Ftaxonomy embeddings |\n| **Sparse Vectors** | BM25\u002FTF-IDF hybrid search | Combine keyword + semantic |\n\n### LLM Runtime\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **ruvllm** | Local LLM inference with GGUF models | Run AI without cloud APIs |\n| **Metal\u002FCUDA\u002FANE** | Hardware acceleration on Mac\u002FNVIDIA\u002FApple | 10-50x faster inference |\n| **[ruvllm-wasm](.\u002Fcrates\u002Fruvllm-wasm)** | Browser WASM runtime: KV cache, HNSW router, MicroLoRA, SONA, chat templates (435 KB) | [`npm`](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@ruvector\u002Fruvllm-wasm) |\n| **RuvLTRA Models** | Pre-trained GGUF for routing & embeddings | \u003C10ms inference → [HuggingFace](https:\u002F\u002Fhuggingface.co\u002Fruv\u002Fruvltra) |\n| **Streaming Tokens** | Real-time token generation | Responsive chat UX |\n| **Quantization** | Q4, Q5, Q8 model support | Run 7B models in 4GB RAM |\n| **π-Quantization (ADR-090)** | 2-bit weights via π-transform + Hadamard rotation + QAT-STE | **10 GB\u002Fs** dequantization, 16x memory reduction |\n| **MoE Memory-Aware Routing (ADR-092)** | Cache-aware expert selection with EMA affinity tracking | **70%+ cache hit rate**, \u003C10µs routing latency |\n\n```bash\nnpm install @ruvector\u002Fruvllm        # Node.js\nnpm install @ruvector\u002Fruvllm-wasm   # Browser (WASM)\ncargo add ruvllm                    # Rust\n```\n\n### Platform & Edge\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **[RVF Cognitive Container](.\u002Fcrates\u002Frvf\u002FREADME.md)** | Single `.rvf` file: store, boot, branch, prove | Replaces Docker + DB + audit system |\n| **rvLite** | Standalone 2MB edge database | IoT, mobile, embedded |\n| **PostgreSQL Extension** | 77+ SQL functions, pgvector replacement | Drop-in upgrade for existing DBs |\n| **MCP Server** | Model Context Protocol integration | AI assistant tool calling |\n| **WASM\u002FBrowser** | Full client-side vector search | Offline-first apps |\n| **Node.js Bindings** | Native napi-rs, zero-copy | No serialization overhead |\n| **HTTP\u002FgRPC Server** | REST API with streaming | Easy microservice integration |\n\n```bash\ncargo install rvf-cli                    # RVF CLI (17 commands)\ncargo add rvf-runtime                    # RVF Rust library\nnpm install @ruvector\u002Frvf                # RVF TypeScript SDK\ndocker pull ruvnet\u002Fruvector-postgres     # PostgreSQL\nnpm install rvlite                       # Edge DB\nnpx ruvector mcp start                   # MCP Server\n```\n\n### Distributed Systems\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **Raft Consensus** | Leader election, log replication | Strong consistency for metadata |\n| **Auto-Sharding** | Consistent hashing, shard migration | Scale to billions of vectors |\n| **Multi-Master Replication** | Write to any node, conflict resolution | High availability, no SPOF |\n| **Snapshots** | Point-in-time backups, incremental | Disaster recovery |\n| **Cluster Metrics** | Prometheus-compatible monitoring | Observability at scale |\n| **Burst Scaling** | 10-50x capacity for traffic spikes | Handle viral moments |\n\n```bash\ncargo add ruvector-raft ruvector-cluster ruvector-replication\n```\n\n### AI & ML\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **Tensor Compression** | f32→f16→PQ8→PQ4→Binary | 2-32x memory reduction |\n| **INT8 CNN Quantization (ADR-091)** | Quantized Conv2D\u002FLinear\u002FPooling with SIMD kernels | **4x memory reduction**, 2x faster CNN inference |\n| **Differentiable Search** | Soft attention k-NN | End-to-end trainable |\n| **Semantic Router** | Route queries to optimal endpoints | Multi-model AI orchestration |\n| **Hybrid Routing** | Keyword-first + embedding fallback | **90% accuracy** for agent routing |\n| **Tiny Dancer** | FastGRNN neural inference | Optimize LLM inference costs |\n| **Adaptive Routing** | Learn optimal routing strategies | Minimize latency, maximize accuracy |\n| **SONA** | Two-tier LoRA + EWC++ + ReasoningBank | Runtime learning without retraining |\n| **Local Embeddings** | 8+ ONNX models built-in | No external API needed |\n| **[Verified Proofs](.\u002Fcrates\u002Fruvector-verified)** | 82-byte proof attestations per vector op | Structural trust, not just assertions |\n| **[Graph Transformers](.\u002Fcrates\u002Fruvector-graph-transformer)** | 8 proof-gated modules: physics, bio, manifold, temporal, economic | Every graph mutation is mathematically verified |\n\n### Specialized Processing\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **SciPix OCR** | LaTeX\u002FMathML from scientific docs | Index research papers |\n| **DAG Workflows** | Self-learning directed acyclic graphs | Complex pipeline orchestration |\n| **Cognitum Gate** | Cognitive AI gateway + TileZero | Unified AI model routing |\n| **FPGA Transformer** | Hardware-accelerated inference | Ultra-low latency serving |\n| **ruQu Quantum** | Quantum error correction via min-cut | Future-proof algorithms |\n| **Mincut-Gated Transformer** | Dynamic attention via graph optimization | **50% compute reduction** |\n| **Sparse Inference** | Efficient sparse matrix operations | 10x faster for sparse data |\n| **Sublinear Solver** | 8 sparse algorithms, O(log n) | Powers coherence, GNN, spectral |\n\n### Self-Learning & Adaptation\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **Self-Learning Hooks** | Q-learning + neural patterns + HNSW | System improves automatically |\n| **ReasoningBank** | Trajectory learning with verdict judgment | Learn from successes\u002Ffailures |\n| **Economy System** | Tokenomics, CRDT-based distributed state | Incentivize agent behavior |\n| **Nervous System** | Event-driven reactive architecture | Real-time adaptation |\n| **Agentic Synthesis** | Multi-agent workflow composition | Emergent problem solving |\n| **EWC++** | Elastic weight consolidation | Prevent catastrophic forgetting |\n\n```bash\nnpx @ruvector\u002Fcli hooks init      # Install self-learning hooks\nnpx @ruvector\u002Fcli hooks install   # Configure for Claude Code\n```\n\n### Attention Mechanisms (`@ruvector\u002Fattention`)\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **46 Mechanisms** | Dot-product, multi-head, flash, linear, sparse, cross-attention, CGT sheaf | Cover all transformer and GNN use cases |\n| **Graph Attention** | RoPE, edge-featured, local-global, neighborhood | Purpose-built for graph neural networks |\n| **Hyperbolic Attention** | Poincaré ball operations, curved-space math | Better embeddings for hierarchical data |\n| **SIMD Optimized** | Native Rust with AVX2\u002FNEON acceleration | 2-10x faster than pure JS |\n| **Streaming & Caching** | Chunk-based processing, KV-cache | Constant memory, 10x faster inference |\n\n> **Documentation**: [Attention Module Docs](.\u002Fcrates\u002Fruvector-attention\u002FREADME.md)\n\n#### Core Attention Mechanisms\n\nStandard attention layers for sequence modeling and transformers.\n\n| Mechanism | Complexity | Memory | Best For |\n|-----------|------------|--------|----------|\n| **DotProductAttention** | O(n²) | O(n²) | Basic attention for small-medium sequences |\n| **MultiHeadAttention** | O(n²·h) | O(n²·h) | BERT, GPT-style transformers |\n| **FlashAttention** | O(n²) | O(n) | Long sequences with limited GPU memory |\n| **LinearAttention** | O(n·d) | O(n·d) | 8K+ token sequences, real-time streaming |\n| **HyperbolicAttention** | O(n²) | O(n²) | Tree-like data: taxonomies, org charts |\n| **MoEAttention** | O(n·k) | O(n·k) | Large models with sparse expert routing |\n\n#### Graph Attention Mechanisms\n\nAttention layers designed for graph-structured data and GNNs.\n\n| Mechanism | Complexity | Best For |\n|-----------|------------|----------|\n| **GraphRoPeAttention** | O(n²) | Position-aware graph transformers |\n| **EdgeFeaturedAttention** | O(n²·e) | Molecules, knowledge graphs with edge data |\n| **DualSpaceAttention** | O(n²) | Hybrid flat + hierarchical embeddings |\n| **LocalGlobalAttention** | O(n·k + n) | 100K+ node graphs, scalable GNNs |\n\n#### Specialized Mechanisms\n\nTask-specific attention variants for efficiency and multi-modal learning.\n\n| Mechanism | Type | Best For |\n|-----------|------|----------|\n| **SparseAttention** | Efficiency | Long docs, low-memory inference |\n| **CrossAttention** | Multi-modal | Image-text, encoder-decoder models |\n| **NeighborhoodAttention** | Graph | Local message passing in GNNs |\n| **HierarchicalAttention** | Structure | Multi-level docs (section → paragraph) |\n| **CGTSheafAttention** | Coherence | Consistency-gated graph transformers |\n\n#### Hyperbolic Math Functions\n\nOperations for Poincaré ball embeddings—curved space that naturally represents hierarchies.\n\n| Function | Description | Use Case |\n|----------|-------------|----------|\n| `expMap(v, c)` | Map to hyperbolic space | Initialize embeddings |\n| `logMap(p, c)` | Map to flat space | Compute gradients |\n| `mobiusAddition(x, y, c)` | Add vectors in curved space | Aggregate features |\n| `poincareDistance(x, y, c)` | Measure hyperbolic distance | Compute similarity |\n| `projectToPoincareBall(p, c)` | Ensure valid coordinates | Prevent numerical errors |\n\n#### Async & Batch Operations\n\nUtilities for high-throughput inference and training optimization.\n\n| Operation | Description | Performance |\n|-----------|-------------|-------------|\n| `asyncBatchCompute()` | Process batches in parallel | 3-5x faster |\n| `streamingAttention()` | Process in chunks | Fixed memory usage |\n| `HardNegativeMiner` | Find hard training examples | Better contrastive learning |\n| `AttentionCache` | Cache key-value pairs | 10x faster inference |\n\n```bash\n# Install attention module\nnpm install @ruvector\u002Fattention\n\n# CLI commands\nnpx ruvector attention list                    # List all 39 mechanisms\nnpx ruvector attention info flash              # Details on FlashAttention\nnpx ruvector attention benchmark               # Performance comparison\nnpx ruvector attention compute -t dot -d 128   # Run attention computation\nnpx ruvector attention hyperbolic -a distance -v \"[0.1,0.2]\" -b \"[0.3,0.4]\"\n```\n\n### Coherence Gate (`prime-radiant`)\n\n| Feature | What It Does | Why It Matters |\n|---------|--------------|----------------|\n| **Sheaf Laplacian** | Measures consistency via E(S) = Σ wₑ · ‖ρᵤ(xᵤ) - ρᵥ(xᵥ)‖² | Mathematical proof of coherence |\n| **Compute Ladder** | Reflex (\u003C1ms) → Retrieval (~10ms) → Heavy (~100ms) → Human | Route by confidence level |\n| **LLM Hallucination Gate** | Block incoherent responses with witnesses | Refuse generation when math says contradiction |\n| **GPU\u002FSIMD Acceleration** | wgpu + AVX-512\u002FNEON + vec4 WGSL kernels | 4-16x speedup on coherence checks |\n| **Governance Audit** | Blake3 hash chai","RuVector 是一个高性能、实时、自学习的向量图神经网络数据库，使用 Rust 构建。其核心功能包括自优化搜索质量、自动调优系统性能以及在本地硬件上运行 AI 模型，无需依赖云 API 且支持多种注意力机制。此外，RuVector 能够跨领域进行知识迁移，提高新任务的学习效率。该系统适用于需要低延迟、高精度向量搜索和图智能处理的应用场景，如复杂数据检索、AI 代理操作系统等，并且能够无缝集成到 PostgreSQL 中或直接在浏览器中运行。",2,"2026-06-11 03:30:47","trending"]