Sieves

Sieves is a library for zero-shot document AI that enables rapid prototyping of document processing pipelines with structured generation and validated output. It bundles common NLP utilities, document parsing, and text chunking with ready-to-use tasks like classification, information extraction, named entity recognition, and more. All organized in an observable pipeline architecture, Sieves is designed for scenarios where structured output is needed but training data is scarce. Sieves is built for developers, data scientists, and AI practitioners who need to quickly build document AI applications without the overhead of training custom models. It is particularly valuable for rapid prototyping and proof-of-concept work, allowing users to chain tasks together in a pipeline using simple Python syntax. The library operates around three key components: Pipeline (orchestrates tasks), Task (pre-built or custom NLP operations), and Doc (core document data structure). ModelWrappers provide backend implementations (Outlines, DSPy, LangChain, etc.), while Bridges connect tasks to model wrappers. This modular design makes Sieves extensible and adaptable to various model backends. What sets Sieves apart is its zero-shot capability: no training data is needed for many tasks. The library also emphasizes structured generation, ensuring model outputs conform to defined schemas. It is maintained by Mantis, an AI consultancy, and is open-source on GitHub.