rUv

High-performance neural network swarm orchestration in WebAssembly

DSPy.ts 2.1 - 100% DSPy Python compliant TypeScript framework with multi-agent orchestration, self-learning capabilities, MIPROv2 optimizer, and comprehensive module library. Powered by AgentDB, ReasoningBank, and Swarm architecture.

SPARC 2.0 - Autonomous Vector Coding Agent + MCP. SPARC 2.0, vectorized AI code analysis, is an intelligent coding agent framework built to automate and streamline software development. It combines secure execution environments, and version control into a

N-API bindings for QuDAG quantum-resistant DAG and cryptography

AI Defense System - Production-ready security middleware for protecting AI applications from prompt injection, manipulation attacks, and adversarial inputs. Available as both 'aidefence' (British) and 'aidefense' (American).

High-performance WebAssembly theorem prover with dependent types, hash-consing (150x faster), Ed25519 proof signatures, MCP support for Claude Code, AgentDB vector search, episodic memory, and ReasoningBank learning. Formal verification with cryptographic

Dual-protocol payment infrastructure for autonomous AI commerce (AP2 + ACP)

WebAssembly-powered temporal analysis toolkit - DTW, LCS, scheduling, and meta-learning

Unified TypeScript API surface for RuVector WASM - attention, learning, economy, and exotic features

Ultra-low latency neural trading engine with Rust + Node.js bindings

⚡ Ultra-fast neural network inference in WebAssembly - sub-microsecond latency

Blazing fast graph pathfinding SDK powered by WebAssembly. 10-15x faster than JavaScript implementations.

V3 Embedding Service - OpenAI, Transformers.js, and Mock providers

Memory module - AgentDB unification, HNSW indexing, vector search, hybrid SQLite+AgentDB backend (ADR-009)

Integration module - agentic-flow@alpha deep integration, ADR-001 compliance

Security module - CVE fixes, input validation, path security

Inflight LLM-stream reformatting — separate a reasoning model's thinking from its answer, collapse repetition, and normalize whitespace, while the tokens are still arriving. Parses Google Vertex / OpenRouter / meta-llm SSE. Rust + WASM (built on ruvnet/mi

Self-learning LLM runtime — TurboQuant KV-cache (6-8x compression), SONA adaptive learning, FlashAttention, speculative decoding, GGUF inference

MRAgent — Cue-Tag-Content graph memory over RuVector, with a Meta-Harness Darwin loop that evolves the reconstruction harness (freeze the model, evolve the harness).

Self-learning LLM runtime — TurboQuant KV-cache (6-8x compression), SONA adaptive learning, FlashAttention, speculative decoding, GGUF inference

HarmonyOS ARM64 native module for @ruvector/ruvllm

Self-learning LLM runtime — TurboQuant KV-cache (6-8x compression), SONA adaptive learning, FlashAttention, speculative decoding, GGUF inference

Multi-LLM Provider System for Claude Flow V3

Shared module - common types, events, utilities, core interfaces

Healthcare optimization with self-learning patient forecasting and swarm-based staff scheduling

Neuromorphic computing with Spiking Neural Networks, STDP learning, and reservoir computing for ultra-low-power ML

Self-learning dynamic pricing with RL optimization and swarm strategy exploration

Self-learning vehicle routing optimization with multi-agent swarm coordination

Self-learning evolutionary game theory with multi-agent tournaments, replicator dynamics, and ESS calculation

Adaptive burst scaling system for ruvector - handles 10-50x traffic spikes

Command-line interface for RuVector vector database with self-learning hooks

NPX package to scaffold new projects with SPARC methodology structure

Quantum-inspired optimization algorithms with swarm-based circuit exploration for combinatorial and constraint problems

Self-learning demand forecasting and swarm-based inventory optimization with uncertainty quantification

Self-learning portfolio optimization with benchmark swarms and multi-objective optimization

Self-learning energy forecasting with conformal prediction and swarm-based ensemble models

Standalone swarm coordination - up to 100+ agents, 4 topologies, hive-mind, consensus

WASM bindings for Neural Trader — market events, coherence gates, replay memory

Performance module - benchmarking, Flash Attention validation, optimization

Command-line interface for QuDAG quantum-resistant DAG operations

Open source Claude Code CLI — based on ruDevolution decompilation of v2.1.91

Testing module - TDD London School framework, test utilities, fixtures, and mock services for V3 Claude-Flow

Pixel-native visual RAG, ported to Rust on the ruvector ANN substrate — screenshot/document retrieval over visual embeddings (a Rust port of PixelRAG) with a metaharness benchmark CLI.

Evaluation + Darwin-Mode bridge for @metaharness/workspace-lens: turn Jacobian-Lens interpretability receipts into a flywheel-consumable workspace_probe score, and reject structurally-brittle prompt/policy mutations (final answer up, workspace grip down).

Jacobian-Lens interpretability primitive for open-weight LLMs (Anthropic 2026-07-06, 'Verbalizable Representations Form a Global Workspace'). Reads the model's functional workspace at inference time — lens_l(h)=unembed(J_l·h) — into workspace tokens, laye

Passive-awareness swarm bus for agent harness pods — threads, non-blocking mentions, and background watchers that fold teammate discoveries in at step boundaries (AgentRadio, arXiv:2607.28430). Freeze the model, evolve the comms policy.

Deterministic long-horizon control with Rust/WASM halt and command guards, real tool execution receipts, full checkpoints, and flush-before-compaction.

The Dream Machine — a config-driven engine for nightly, cloud-scheduled, evidence-gated repository evolution. It never merges; a human does.

Governed autonomous variation runtime: evidence-driven repository engineering loops with bounded authority, replayable receipts, semantic supervision, and RVF working memory.

Radio real-time streaming peer expert mesh — bounded, signed, constitution-pinned distributed MoE. @metaharness/radio is the metadata-only control plane; a signed direct peer transport carries expert streams. True logit mixing where tokenizers match; safe

A verifiable self-improvement loop for agent harnesses. Freeze the model. Evolve the harness. Promote only what proves lift.