agrag.chunking.rules.DEFAULT_CHUNKING

DEFAULT_CHUNKING = Chunking(rules=(ChunkingRule(match=RuleMatch(loader_names=['docling']), chunker=DoclingChunker()),), fallback=RecursiveChunker(tokenizer='character', chunk_size=1024, min_characters_per_chunk=24))

The preset that Graph uses: docling loads go to DoclingChunker.

All other documents go to a RecursiveChunker with a 1024-character size.