Cleora

Cleora is a graph embedding engine that operates deterministically on CPU. It aims at producing production-grade embeddings from graphs by avoiding sampling and training steps typical of other methods. The tool targets researchers and engineers who require reproducible graph representations and fast embedding generation on commodity hardware.

Cleora uses a sparse Markov matrix constructed from the input graph and applies matrix powers to capture the full distribution of all walks. Each iteration propagates embeddings through neighbors with L2 normalization, ensuring stable embeddings. No GPU or negative sampling is involved, and the approach emphasizes determinism and exact walk distributions.

Embeddings produced by Cleora are dense vectors for every entity, suitable for downstream ML tasks such as classification, clustering, and similarity search. The library provides a unified API for multiple embedding algorithms, evaluation metrics, and analysis tools, all within a compact, dependency-light package.

Cleora documents plans and capabilities around CPU-only execution, heterogeneous and hypergraph support, and integration pathways via a CLI and Python API. The project emphasizes openness, small footprint, and compatibility with common data science stacks, while noting its absence of GPU requirements and CUDA dependencies.