ruv-FANN: The Neural Intelligence Framework 🧠

What if intelligence could be ephemeral, composable, and surgically precise?

Welcome to ruv-FANN, a comprehensive neural intelligence framework that reimagines how we build, deploy, and orchestrate artificial intelligence. This repository contains three groundbreaking projects that work together to deliver unprecedented performance in neural computing, forecasting, and multi-agent orchestration.

🌟 The Vision

We believe AI should be:

• Ephemeral: Spin up intelligence when needed, dissolve when done
• Accessible: CPU-native, GPU-optional - built for the GPU-poor
• Composable: Mix and match neural architectures like LEGO blocks
• Precise: Tiny, purpose-built brains for specific tasks

This isn't about calling a model API. This is about instantiating intelligence.

🎯 What's in This Repository?

1. ruv-FANN Core - The Foundation

A complete Rust rewrite of the legendary FANN (Fast Artificial Neural Network) library. Zero unsafe code, blazing performance, and full compatibility with decades of proven neural network algorithms.

2. Neuro-Divergent - Advanced Neural Forecasting

27+ state-of-the-art forecasting models (LSTM, N-BEATS, Transformers) with 100% Python NeuralForecast compatibility. 2-4x faster, 25-35% less memory.

3. ruv-swarm - Ephemeral Swarm Intelligence

The crown jewel. Achieves 84.8% SWE-Bench solve rate, outperforming Claude 3.7 by 14.5 points. Spin up lightweight neural networks that exist just long enough to solve problems.

🚀 Quick Install ruv-swarm

# NPX - No installation required!
npx ruv-swarm@latest init --claude

# NPM - Global installation
npm install -g ruv-swarm

# Cargo - For Rust developers
cargo install ruv-swarm-cli

That's it. You're now running distributed neural intelligence.

🧠 How It Works

The Magic of Ephemeral Intelligence

1. Instantiation: Neural networks are created on-demand for specific tasks
2. Specialization: Each network is purpose-built with just enough neurons
3. Execution: Networks solve their task using CPU-native WASM
4. Dissolution: Networks disappear after completion, no resource waste

Architecture Overview

┌─────────────────────────────────────────────┐
│          Claude Code / Your App             │
├─────────────────────────────────────────────┤
│            ruv-swarm (MCP/CLI)              │
├─────────────────────────────────────────────┤
│         Neuro-Divergent Models              │
│    (LSTM, TCN, N-BEATS, Transformers)      │
├─────────────────────────────────────────────┤
│           ruv-FANN Core Engine              │
│        (Rust Neural Networks)               │
├─────────────────────────────────────────────┤
│            WASM Runtime                     │
│    (Browser/Edge/Server/Embedded)          │
└─────────────────────────────────────────────┘

⚡ Key Features

🏃 Performance

• <100ms decisions - Complex reasoning in milliseconds
• 84.8% SWE-Bench - Best-in-class problem solving
• 2.8-4.4x faster - Than traditional frameworks
• 32.3% less tokens - Cost-efficient intelligence

🛠️ Technology

• Pure Rust - Memory safe, zero panics
• WebAssembly - Run anywhere: browser to RISC-V
• CPU-native - No CUDA, no GPU required
• MCP Integration - Native Claude Code support

🧬 Intelligence Models

• 27+ Neural Architectures - From MLP to Transformers
• 5 Swarm Topologies - Mesh, ring, hierarchical, star, custom
• 7 Cognitive Patterns - Convergent, diverge

... View full README on GitHub