Hezar is an all-in-one AI library for the Persian language, gathering models and tools for Persian NLP, speech recognition, OCR, and image captioning behind one developer-friendly interface. Its name means “thousand” in Persian. The project is Apache-2.0 licensed, has earned over 1,000 stars on GitHub, and remains actively maintained, with updates during 2026.
Working with Persian AI usually means hunting down scattered models, each with its own incompatible API and setup quirks. Hezar fixes that by unifying the best available works for Persian under a task-based interface: load a model for text classification, speech recognition, optical character recognition, or image captioning in a couple of lines of code, and let the library handle the plumbing. It integrates seamlessly with Hugging Face Hub, where its models live, and bundles the supporting cast โ tokenizers, word embeddings, and feature extractors โ plus supplementary tools for deployment, benchmarking, and optimization, so taking a Persian model from notebook to production stays within one coherent toolkit.
Hezar also pays attention to the unglamorous parts of applied AI: consistent preprocessing, sensible defaults, and documentation aimed at practitioners rather than only researchers. For product teams, that translates into shorter paths to features like Persian voice input, document digitization through OCR, or automatic captioning for Persian media. By gathering models that were previously scattered across individual researchers’ repositories into one maintained library with a shared interface, Hezar lowers the risk of building on Persian AI tooling that might otherwise be abandoned or incompatible.
Hezar is free and open source under the Apache-2.0 license; its models are hosted openly on Hugging Face Hub.
Hezar is built for developers and researchers serving Persian-speaking users who need production-ready Persian language and speech capabilities without assembling them from scratch.