// the find
ruvnet/ruv-FANN
A blazing-fast, memory-safe neural network library for Rust that brings the power of FANN to the modern world.
A Rust rewrite of FANN, the C library for multilayer neural networks, and the piece of this monorepo a Rust developer would actually add to Cargo.toml. The README leads with ruv-swarm, an npm and MCP agent-orchestration tool, and also covers a forecasting project (Neuro-Divergent), so the repo reads more like a platform pitch than a library page. It suits someone who wants a small CPU-only network crate and is willing to read the source instead of trusting the README.
Porting FANN gives the core a fixed reference point: the algorithms and training behaviour come from the original, so the crate can be checked against a known implementation rather than an invented API. The dual Apache-2.0 and MIT licensing is the standard Rust arrangement and raises no licensing question for most downstream users. The CPU-only design with a WASM target means small models can run in a browser or on an edge device without a CUDA dependency.
The headline numbers (84.8% SWE-Bench, 2.8-4.4x speedups, 29% less memory) come from ruv-swarm and the forecasting models, with no hardware, baseline version, or reproduction steps. None of them tell you how fast the FANN core is. There is no code sample for the core crate: the install section shows npx ruv-swarm, and the core API is only reachable through docs.rs, so a reader cannot tell how to build, train, or save a network from the README. 'Zero unsafe code' and 'zero panics' are stated with no enforcement mechanism shown; indexing and integer overflow can panic in safe Rust, and the README offers neither a forbid-unsafe attribute, a clippy lint, nor a test claim to back the second one. The acknowledgments credit contributors with replacing mock implementations and removing mock functions from the WASM bindings, which is an honest credit but means the code's maturity is recent, and the README gives no test coverage or stability statement.