what is local-recall fr?

Local Recall is an early prototype that works like Microsoft's Windows Recall feature, but keeps everything on your own computer instead of sending anything to the cloud. It periodically takes screenshots of your screen, reads the text out of them, and lets you later search through your history by meaning, or ask a local AI assistant questions about what you were looking at. The README is upfront that this is an early proof of concept still under active development, with things like the database structure likely to change. The pipeline works in stages. A capture loop takes a screenshot every five seconds and compares it to the previous one using a perceptual hash, skipping duplicates so unchanged screens do not pile up in storage. A background worker then reads text out of new screenshots using Windows' built in OCR engine. Another background worker turns that extracted text into numerical representations, called embeddings, using a small local language model, and stores them in a SQLite database extended with a vector search add on, which allows fast similarity based searches over your history. A locally run large language model, connected through a tool called LM Studio, uses those search results to answer questions you type about your past screen activity. Because it relies on Windows' built in OCR and some Windows specific system calls, it only runs on Windows 10 or 11. Setting it up involves installing LM Studio and loading a language model in it, creating a Python virtual environment, installing the pinned dependencies, and editing a configuration file to set things like how often screenshots are taken and which local model to use. Once running, you can search your history from the command line, chat with an assistant about it in a terminal, or use a simple local web page that shows matching screenshots. The README does not mention a software license.