pihme/ragite v0.3.0 was released on October 16, 2026. ragite is a retrieval (RAG) tool that is embedded and used inside a database in the form of a SQLite extension, and it aims to run on CPU. This release changes the default retrieval model and moves the large file of the tokenizer — the component that splits text into processing units — into the repository, affecting compatibility. The basis is the ragite/v0.3.0 release notes on GitHub.
Besides the default retrieval model change and the tokenizer large file move, the release notes list three more changes. A problem with assembling model weights (model weight, the numeric values a model has learned) on Mac and Windows was fixed, and table cell boundary handling was also corrected. A feature was added for finding chunks (chunk, a small piece into which a document is split) by row identifier. The configuration of performing retrieval inside SQLite means that retrieval finishes within the database, without standing up a separate retrieval-only service.
However, the only thing that can be verified is the published release notes. The license and the primary language are not in the material, and there are no measurements showing how the result of changing the default retrieval model shows up in retrieval accuracy or processing speed. The repository's star count is aggregated as 0. Because the retrieval model and the tokenizer file location changed together, those applying 0.3.0 need to check existing retrieval results and deployment configuration again. The impact of the row identifier feature on actual retrieval is difficult to gauge from this record alone.