Feature
Document chunking
FAQai splits each document into chunks before it generates Q&A and evaluation data. The chunk is the retrieval unit you export. FAQai does not embed it for you.
Strategies that exist
- Heading-aware is the default. It keeps heading structure and overlaps neighbouring text.
- Fixed size cuts uniform chunks and ignores heading structure.
- Paragraph splits on blank lines and packs paragraphs up to the target size. A paragraph longer than the maximum is still split.
These are the implemented strategies. FAQai does not offer a separate semantic, hierarchical, recursive, or agentic chunker.
Presets
- Balanced, the default: about 400 tokens, 1,600 target characters, 2,400 maximum, 250 characters of overlap.
- Fine-grained: about 200 tokens, 800 target characters.
- Large context: about 800 tokens, 3,200 target characters.
The Free plan uses Balanced. Paid plans can set a custom target from 128 to 2,048 tokens, overlap up to 25 percent, and one of the three strategies. Paid plans can also add a context header and hypothetical questions to each chunk. That enrichment is optional and uses a separate monthly quota.
Read the chunking strategy guide before changing the preset.