What Is Megachunking?

Megachunking is Morphos AI's patent-pending method for representing documents as a hierarchy of semantically coherent chunks rather than fixed-size windows. It eliminates the chunk-size tradeoff in conventional RAG by letting retrieval surface the right level of context for each query, instead of forcing a single fixed chunk size for every case.

Why chunking matters in retrieval

RAG systems divide documents into chunks before vectorizing them. Traditional approaches use fixed character or token lengths, which often split related ideas across chunk boundaries and force the same chunk size on every query, whether the query needs a sentence or a section. This fragments meaning and degrades retrieval quality.

How megachunking improves on fixed-size chunking

Megachunking represents each document as a hierarchy of semantically coherent chunks rather than a single fixed-size split. Retrieval can then surface the right level of context for each query, from a tight passage to a broader section, instead of being locked into one chunk size for every case.

Where megachunking fits

Megachunking is one of three named patent-pending innovations within Green Vectors, alongside continuous vectorization and auto weighting. It is delivered through Kitana.