Configurable Chunking

What this page is for

Control how ingested content is split into chunks using LLM-powered or character-based methods.

Chunking methods

LLM chunking (default)

Uses OpenAI to intelligently split content into semantically coherent chunks. The LLM identifies natural boundaries based on topic, structure, and meaning.

• Best for: prose, documentation, articles
• Produces: typed chunks (knowledge, navigation, table_row, glossary, faq, code)
• Extracts: heading paths, tags, language detection

Character chunking

Splits content by character count with paragraph and sentence boundary detection. Uses configurable chunk size and overlap.

• Best for: large documents where LLM cost is a concern
• Produces: uniform-sized text chunks
• Configurable: chunk size (100–10,000 chars), overlap (0–2,000 chars)

Configuration

Pass chunkingConfig in any ingest request body:

{
  "method": "llm",
  "chunkSize": 1000,
  "overlap": 200
}

Field	Type	Default	Description
method	"llm" | "character"	"llm"	Chunking strategy
chunkSize	number (100–10,000)	1000	Target chunk size in characters
overlap	number (0–2,000)	200	Overlap between consecutive chunks

Examples

LLM chunking (default)

curl -X POST http://localhost:3000/api/ingest/manual \
  -H "Content-Type: application/json" \
  -H "X-User-ID: user-1" \
  -d '{
    "content": "Your document text here..."
  }'

Character chunking with custom settings

curl -X POST http://localhost:3000/api/ingest/manual \
  -H "Content-Type: application/json" \
  -H "X-User-ID: user-1" \
  -d '{
    "content": "Your document text here...",
    "chunkingConfig": {
      "method": "character",
      "chunkSize": 500,
      "overlap": 100
    }
  }'

Verify

• After ingestion, GET /api/session/<id> shows generated chunks.
• LLM chunks have typed chunkType values and extracted metadata.
• Character chunks are uniformly sized with overlap at boundaries.

Next steps

• Sessions — review and edit generated chunks.
• Publishing — publish curated chunks to a collection.