amazinghorseli/rednote-insight

RedNote Insight is a Python tool for analyzing product reviews from Xiaohongshu, the Chinese social platform that combines Instagram-style posts with e-commerce. It is aimed at people who sell or source products and want to find out what customers actually complain about or ask for in a given product category, without reading hundreds of posts by hand. The tool has two modes. In question-and-answer mode, you ask a natural-language question like "which brand of magnetic sensor lights gets the best reviews?" and the system searches its internal knowledge base of posts to give you a structured answer with pros and cons per brand. In insight mode, you type a product category name and the tool produces a structured market report: the top three user complaints (each with an estimated frequency), unmet needs, competitor positioning, and a scoring of the market opportunity. If the knowledge base does not yet contain data for the category you typed, the tool automatically generates a set of representative posts using the AI model and adds them to the database before running the analysis. This means you are not limited to the categories already loaded. Under the hood the tool stores posts in a local vector database (ChromaDB) and uses a combination of keyword search and semantic similarity to find relevant content, then reranks the results with a second AI model before passing them to the language model. Multiple specialized AI agents handle different steps: one routes the query, one analyzes comments, one aggregates patterns, and one writes the final report. All of this runs in a single Python process with no need for Docker, a separate database server, or any other external service. You can load your own data by importing a CSV or Excel file of posts. The app runs in a web browser via Streamlit. It supports any AI API provider that is compatible with the OpenAI format, including DeepSeek, SiliconFlow, and others. The README is written primarily in Chinese.