How ruvnet/ruv-fann works
Project type and purpose
• This is a multi-module, full-stack monorepo for a Rust-based neural-network foundation (ruv-FANN), an advanced forecasting library (neuro-divergent), and a multi-agent orchestration framework (ruv-swarm) with WebAssembly and JavaScript/Node.js front-ends.
• The README emphasizes:
• ruv-FANN: core Rust crate implementing FANN-compatible neural networks
• neuro-divergent: Rust forecasting library built on ruv-FANN
• ruv-swarm: multi-agent system using ruv-FANN and hosting rich CLI, WASM, and npm bindings
File structure
• Top-level Rust crates under Cargo.toml:
• src/… (core ruv-FANN code)
• neuro-divergent/…
• ruv-swarm/…
• Under ruv-swarm/crates: many sub-crates—for core swarm logic, transport, persistence, ML, DAA, MCP, CLI, WASM builds, benchmarks
• Under ruv-swarm/npm: full JS/TypeScript wrapper, tests, Dockerfiles, build scripts
• Under claude-code-flow, daa-swarm, and other plan directories: design documents and planning
• Standard CI configs, Docker compositions, tests and examples interwoven
README for insights
• Core architecture: layered Rust crates → WASM targets → npm package → CLI and browser examples
• Highlights dependencies: polars, serde, rayon, wasm-bindgen, Docker, Jest, CI pipelines
Detail
• Link CLI commands → internal core functions
• End-to-end stack: data → neural network → forecast → swarm orchestration → client
• Don’t omit the I/O subsystem (binary, JSON, FANN formats) or training algorithms (Backprop, RPROP, Cascade)