Framer AI AgentsDesign and publish professional sites with AI
Linkwise
Save, annotate, chat with everything you've read.
9 followers
Save, annotate, chat with everything you've read.
9 followers
Linkwise lets you save articles, newsletters, RSS, emails, and PDFs - and turns them into a knowledge base you can chat with. Instant insights, quizzes, flashcards, mindmaps, and time-synced YouTube transcripts. The Pocket replacement built for people who actually read.
How well does the chat handle really long PDFs with mixed text and charts, does it cite back to the page or section it pulled from?
Hey @ceylinegil91932! To keep things incredibly fast and efficient right now, Linkwise handles long PDFs by extracting and storing the top 10 most text-dense, high-value chunks from the document upon import. This gives the AI the core substance and key sections of the file to chat with, without bloating your library. I’m definitely planning to increase this chunk limit in the future to support even deeper coverage for massive files!
For charts, it relies on the text and data it can cleanly extract from those key sections. As for citations: right now, the chat will point you directly to the specific document it's pulling from so you know exactly which file the knowledge belongs to. Precise page-level linking isn't live just yet, but deepening those reference tools is absolutely on the roadmap as I keep refining the document experience!
How does the chat with your knowledge base actually work under the hood, is it pulling answers straight from your saved items or leaning on a general model and risking made up stuff?
Hey @ayhan373792! Great question. To prevent the AI from just making things up or hallucinating, we use an architecture called RAG (Retrieval-Augmented Generation).
Under the hood, when you ask a question, we don't just pass it blindly to a general AI model. First, we run a semantic search across the documents or collections you’ve selected to find the exact, relevant paragraphs. Then, we feed only that specific context into the LLM, instructing it to answer your question strictly using your saved data. If the answer isn't in your library, it will tell you, rather than leaning on general knowledge and risking made-up information. Your data acts as the absolute source of truth!
The chat with your knowledge base sounds genuinely useful, especially the quiz and flashcard generation. One thing that would make it a real Pocket replacement for me is a solid web clipper that grabs the full article text cleanly, even from sites with paywalls or heavy JS. Most clippers fall apart on those.
@ensar90917 thanks! You hit the exact pain point, heavy JS and paywalls are where most clippers break. I’ve actually been working extensively on our backend parsing technology to handle exactly this, stripping out the client-side bloat so the AI gets completely clean text. Paywalls are always a cat-and-mouse game, but optimizing for these tough sites is a major focus. I'd love for you to give it a spin and throw your hardest "problem sites" at it!