Towhee

Towhee is an open-source machine learning pipeline that encodes unstructured data — images, text, audio, molecular structures, and nearly 20 data modalities in total — into embeddings, backed by 700+ pre-trained models spanning computer vision, NLP, multimodal, audio, and medical domains. It provides end-to-end pipeline optimizations for performance, a Pythonic API with method-chaining for building pipelines readably, and integrates with popular ML libraries and vector databases (a natural pairing with tools like Milvus). Free and open source, hosted on GitHub with active documentation and community engagement. Good fit for a developer who needs to turn diverse, unstructured data into embeddings for a vector search or RAG pipeline, and wants a large library of pre-trained models rather than training embedding models from scratch.