wtf is instagram-local-rag?

instagram-local-rag turns any public Instagram account into a searchable knowledge base that runs entirely on your own computer, with no cloud service and no per-use billing. The idea is that a lot of genuinely useful information lives inside Instagram carousels and reels, but it is hard to search, mixed in with promotional posts and memes, and locked inside images and audio rather than text. This project pulls that content out and makes it something you can ask questions against, with the answers pointing back to the original posts. The pipeline runs in stages. First it scrapes captions and post information from the target account. A local AI model then filters out noise like promotional posts, keeping only the content that matches categories you define. For posts that survive filtering, the tool downloads the actual videos and images, transcribes video audio into text, and uses a vision model to read text out of image slides. All of the extracted text is then merged, split into chunks, and turned into a searchable index. Finally, a question and answer script retrieves the most relevant chunks and produces an answer grounded only in that retrieved text, always citing which posts it came from. Everything runs locally through Ollama, a tool for running AI models on your own machine, along with a speech to text library called faster-whisper that works fine on a regular CPU. Authentication to Instagram works by reusing a cookie from your own browser's existing login, rather than typing or storing a password anywhere. The filtering categories are fully customizable through a plain text prompt file, so the same pipeline can be pointed at any account and topic, not just the study abroad advice account the author originally built it for. The README is upfront that local AI models are good at pulling information out of text but weaker at reasoning over it, so the question and answer step is intentionally strict about only using retrieved context rather than inventing answers. It also includes clear guidance that the tool is meant for personal research on public content, that scraped data should never be shared or committed to the repository, and that Instagram's terms of service and rate limits should be respected. The project is released under the MIT license.