ThinkRAG

ThinkRAG answers questions about your own documents using a knowledge base on your laptop. It's for researchers, students and other knowledge workers who want to find information across their files, with particular attention to Chinese text.

• Document collection: Add PDF, DOCX and PPTX files or content from web page URLs, then browse the collection and delete documents you no longer need.
• Answers with references: Questions search your knowledge base and return answers alongside retrieved reference documents, using hybrid search and reranking to select relevant material.
• Chinese text support: Chinese character splitting, title enhancement and Chinese prompts support document processing and answers, with bilingual BAAI BGE models for retrieval.
• Local model choice: Ollama runs models such as DeepSeek, Qwen and Gemma locally, while embedding and reranking support includes models from Hugging Face.

ThinkRAG runs on macOS and Linux through a browser interface and doesn't require a GPU. Windows has unresolved compatibility issues. Offline use requires downloading the language, embedding and reranking models first. Knowledge base data stays on your computer, and development mode uses local file storage without a separate database.

You can also use OpenAI or compatible APIs, including DeepSeek, Moonshot and ZhiPu. Those connections require internet access and API credentials, and send model requests to the chosen provider. ThinkRAG is open source under the MIT license.