The knowledge chunk splitter uses the same AI splitting method as the knowledge base import to divide your content into complete knowledge blocks. You can paste text directly into the input box, or upload txt, docx, xlsx, pptx, or pdf files — the system first extracts the plain text from each file before processing. After you press send, the screen streams the splitting progress: the AI first scans the whole document to build a chapter structure map, then splits by priority — one chapter per block where possible, one paragraph per block when there are no chapters, one concept per block when there are no paragraphs, and about 2,000 characters per block when there are no clear concepts — and automatically generates search keywords that include the book title and chapter names. When splitting is done, the screen shows all knowledge blocks in JSONL format identical to the downloadable file, and a JSONL file is generated with a download link. Each line is one knowledge block record with output (the original text of the block), keyword, remark (x/y, marking which part of the content it is out of the total), and vector (the block content encoded into an embedding by an AI vector model, ready for vector search) fields. The resulting JSONL file can be imported into a knowledge base directly with the knowledge base's Import JSON feature, or used for RAG retrieval and fine-tuning data preparation. Long content is analyzed window by window, so please wait for the stream to finish. Prompt example 1: (Upload a company employee handbook docx and send it as is) Prompt example 2: (Upload a course handout pptx and an FAQ xlsx together and send them as is) Prompt example 3: Chapter 1 Leave Policy: Request personal leave one day ahead; sick leave needs a doctor's note. Chapter 2 Overtime: Weekday overtime needs manager approval. Prompt example 4: (Upload a product user manual pdf and send it as is) Prompt example 5: (Upload a txt collection of hospital patient education articles and send it as is) Prompt example 6: (Upload chapter three of a university textbook as a pdf and send it as is) Prompt example 7: (Upload an xlsx spec comparison table of all product models and send it as is) Prompt example 8: (Upload three company annual reports from different years as pdf files together and send them as is) Prompt example 9: The community garbage truck reaches the courtyard at 7 p.m. Monday to Friday. Sort recycling into paper, plastic and metal. Book bulky furniture pickup with the management office in advance. Prompt example 10: Machine learning lets computers learn patterns from data. Deep learning is a branch of machine learning that uses multi-layer neural networks. Transfer learning applies learned knowledge to new tasks. Maximum total upload size per request: 512MB (all files and message content combined). Accepted document formats: pdf, docx, xlsx, pptx, txt, csv (file content is provided to the model as extracted text). Accepted image formats: gif, jpeg, jpg, png, webp. Animated GIF images are analyzed using only the first frame. Maximum number of images: 600. When uploading more than 20 images, each side must be at most 2,000 pixels. Maximum size per image: 10MB (base64-encoded). Maximum total image size per message: 20MB (text and history share the same request size limit; uploads close to the limit may still be rejected). Image dimensions: 200 to 8,000 pixels per side. Output content formats: Text Pricing: Medium-high Maximum input per run: about 5,000,000 tokens. File output format: JSONL (importable into the knowledge base). Continuous conversation: Not supported AI can make mistakes. Please verify important information.