Interface: IMdChunkerConfig
Defined in: internal/md-chunker/types.ts:4
Configuration of the MdChunker. Lands in IAIGatewayConfig.mdChunker.
Properties
maxTokens?
optionalmaxTokens?:number
Defined in: internal/md-chunker/types.ts:18
The target size of a chunk, in estimated tokens (characters / 4).
Big enough to hold one idea, small enough that several fit the history budget next to the answer.
A section over it is cut at paragraph, line, then sentence borders; a fenced code block is never cut, so a chunk that holds one may be larger.
Examples
600 - a few paragraphs: the default for prose documentation
300 - FAQ-style documents with short, self-contained answers
Default
600
minTokens?
optionalminTokens?:number
Defined in: internal/md-chunker/types.ts:24
Below this a section merges with its neighbour, so a heading with two lines is not a chunk of its own.
Default
100
overlapTokens?
optionaloverlapTokens?:number
Defined in: internal/md-chunker/types.ts:31
The overlap between consecutive pieces of one cut section: the last units of a piece open the next one, so a sentence on the border is whole in one of them.
Default
80