Kind App by Synsira - A new offline pdfCHAT type service.
I'm not sure if some of you are still searching for ways to efficiently chat with PDFs and other documents. Some of the local offline ChatGPT solutions that were hyped a couple of years ago have either disappeared or no longer receive dev support, and they don't support new models. I've tried Lmstudio and Ollama, but their quality of response hasn't met my expectations. That's when Google Chat (the one that's now available on Google search by default) suggested I try Kind App for its privacy features, which are up-to-date and might work for my use case. However, there's almost no information available about this AI tool, I couldn't find anything online. I decided to take a chance and installed the app anyway. Kind App offers two service options: a free plan and a paid plan for $18 per month. The free plan allows for three collections, each of which can hold up to 100 PDFs, as far as I understand. I uploaded around 20 academic books, and it took some time to extract text, chunk, and create a database. One of those books has almost 800 pages. At this point, the local ChatGPT options would start to hallucinate, they'd make things up if I asked a question, and sometimes they wouldn't even provide the correct answers. As a result, I only used them on a per-PDF session basis; while the answers were still not as reliable as those from online ChatGPT, it was an acceptable workaround for me. But what really surprised me was how accurate the new Kind AI tool was. I randomly checked a page from one of the books and asked it a question about the subject – and it quickly provided the accurate information. It excelled in so many inquiries I made. Another neat feature of the tool is that it doesn't hallucinate if it doesn't know the answer; instead, it simply says so. I did some digging into the tool and read through their fineprint on their website. From what I understand, they're using Gemini Pro in the backend to handle the prompt queries, and they claim that no intellectual property will be uploaded to any cloud server – instead, it will be stored locally on the user's disk. So, essentially, all the books I uploaded and their chunks are stored on my disk, and it's only the query part that's being handled by Gemini Pro. As someone relatively new to the AI scene, I'm wondering, is it safe to use Kind App without worrying about my privacy or the books being used for AI training by Google?
