Chunking

Understand how Orka AI splits documents into optimally-sized chunks for embedding and retrieval.

How Chunking Works

Orka AI uses a recursive text splitter that intelligently cuts text at natural boundaries:

Try paragraph breaks (\n\n) first

Fall back to line breaks (\n)

Then sentence boundaries (. )

Then word boundaries ( )

Finally, character-by-character

Configuration

await orka.knowledge.create({
  name: 'docs',
  source: myContent,
  chunkSize: 1000,     // Max characters per chunk
  chunkOverlap: 200,   // Overlap between consecutive chunks
});

Recommended Sizes

Content Type	Chunk Size	Overlap	Why
FAQ / Q&A	300–500	50–100	Each Q&A is self-contained
Technical docs	800–1200	150–250	Preserves code blocks
Long articles	1000–1500	200–300	Balances context & specificity
Legal / contracts	500–800	200	Precise clause retrieval

💡 Why Overlap?

Overlap ensures information at chunk boundaries isn't lost. When a chunk ends mid-sentence, the next chunk starts a few hundred characters earlier, capturing the full context.