Gemini Notebook (formerly NotebookLM — Google renamed it in July 2026; same product, same free tier) is the most polished example of source-grounded AI. The product is built around a strict idea: only answer from the sources I uploaded, cite every claim, never hallucinate from world knowledge. Its Studio panel has grown a lot: Audio Overviews in four formats and 80+ languages, Video Overviews, slide decks, infographics, mind maps, flashcards, quizzes, Data Tables that export to Sheets, and a cloud code runtime for data analysis. For synthesizing a stack of papers or summarizing a course pack, it's genuinely excellent.
But the strict philosophy is still its ceiling. You can't pick a model. You can't toggle thinking. The flashcards have no spaced-repetition schedule and no export; the quizzes aren't a graded, timed exam; there's no study schedule, no live recording, no essay or lab-report writer, and no team library — sharing is still per notebook. Video Overviews are short summaries, not a 45-minute lecture. Once you move from "understand what I uploaded" to "get me through this exam, this certification, or this training rollout," you have to leave the tool.
Scholarly takes the same source-grounded core — upload a PDF, get accurate answers with citations, generate an audio overview — and treats it as one feature inside a full study platform. You can swap between GPT, Gemini, Grok, and Claude. You can tune every generation with custom instructions and presets. And once the AI has helped you understand the material, you keep going: SM-2 spaced repetition, graded practice exams, chaptered video lectures up to 45 minutes, three-speaker podcasts up to 60 minutes, and a shared team library. Same notebook idea, with a much bigger ceiling.