SimplyRetrieve
A fully local, GUI and API platform for Retrieval-Centric Generation that lets a user build a private chat tool over their own documents and a locally hosted open-source language model, with a retrieval-tuning module for prompts and analysis.
SimplyRetrieve provides a Gradio-based GUI and API for Retrieval-Centric Generation, an approach the project positions as building on Retrieval-Augmented Generation by emphasizing the language model's role in interpreting context while the retriever component handles knowledge memorization, aiming for more efficient and interpretable generation with smaller models. Its retrieval-tuning module covers prompt engineering, tool configuration, and retrieval analysis, and a private knowledge-base constructor lets a user build a custom document collection to query on the fly through a dedicated knowledge tab.
It runs on a single NVIDIA GPU such as a T4, V100, or A100, is built on Hugging Face, Gradio, PyTorch, and Faiss, and supports multi-user concurrent access through Gradio's queue mechanism. Its default configuration uses an instruction-tuned, uncensored open model and a multilingual embedding model, though the project reports it works with many other open-source LLM and retriever sizes, including non-English languages with appropriate prompt templates.
A technical report and short video walkthrough of the tool are linked from the project.