Yaadsi

We all save Reels, carousels, posts and articles "for later". Later never comes. I checked my own Saved folder: 71 links saved to learn from, almost none opened again. 72% were Instagram carousels and Reels, knowledge locked inside images and video that read-later apps can't read. And 1 in 6 had already vanished a few months later, deleted or made private. Yaadsi turns the links you save and forget into a short daily lesson, so what you saved becomes what you know. You keep saving the way you already do. Yaad, a small study buddy, does the rest: Reads everything you save, even images and video: every carousel slide by OCR, and a short brief for each Reel. Each save is filed by topic, time to read and difficulty. Plans your backlog for you. The Today screen shows the pile in five numbers (Saved, Learned, Waiting, Fading, Gone) and picks a coherent 15-minute bite each day. There's nothing to set up. Teaches it at your level, from your own saves, as a short article you'd actually read. It has a hook, one analogy, a takeaway, the real diagram cropped from the post you saved, a simple picture of the idea, and a quick quiz. One tap changes the depth or the format. Checks you remember, right before you'd forget, with spaced review. Finds things from half-memory, in words or by voice: "that Reel about writing video prompts as JSON". Connects the dots across everything you saved on a topic: what agrees, what disagrees, what's missing, and the order to learn it in. Each Meta model has one job. Muse Spark Open is the brain: it writes the lessons, plans Today, answers questions and reads new links. Its live reasoning streams as "Yaad is thinking". All your saves fit in one cached prompt of about 39k tokens, so asking across everything costs about a cent. Muse Wonder judges three drafts of every lesson and picks the best. SAM 3.1 crops the creator's real diagram out of the slide. Muse Image draws the picture, and Muse Voice Transcribe lets you ask out loud. It's measured, not just demoed. Reading the slides took my Instagram saves from 55% to 80% understandable. Ask found 12 of 12 half-remembered posts. In a blind test I preferred the judged lessons 6 to 2. A judged lesson costs about 4 cents. Try it live at yaadsi.vercel.app, or run the public repo with no keys in one minute. Yaad means memory in Hindi. Yaadsi remembers, so you don't have to.