Split PDF text into overlapping token-sized chunks for embeddings and RAG. PDF Chunker for RAG is a free online tool from the ToolsRift PDF Tools collection. It runs entirely in your browser — nothing to install, no signup, and your data never leaves your device.
Upload a PDF and set your chunk size and overlap in approximate tokens. The tool rebuilds paragraphs from the PDF, packs them into chunks that stay under your size limit, and carries the tail of each chunk into the next so context is never cut mid-idea. Export as .json or .jsonl.
Embedding models have a fixed context window, and retrieval quality drops when a chunk mixes unrelated topics. Chunking on paragraph boundaries with a small overlap keeps each chunk coherent and retrievable.
Using the standard rough estimate of about four characters per token for English prose. It is an approximation — verify against your model's own tokenizer if you are close to a hard limit.
512 tokens with 50 tokens of overlap is a solid default for most embedding models. Use smaller chunks for precise fact retrieval and larger ones when answers need more surrounding context.
JSON gives one array of chunk objects. JSONL puts one chunk per line, which streams better and is what most vector-database loaders expect.