CodeChunk

A CodeChunk is the atomic unit of the index. Every chunk represents a single logical unit of code or documentation extracted from a source file.

Fields

Field	Description
id	Stable 16-char hex hash of projectId + filePath + startLine
projectId	Isolates chunks belonging to this project
filePath	Path relative to projectRoot
language	Any language supported by @kreuzberg/tree-sitter-language-pack (e.g. typescript, python, rust, go, …)
type	See Chunk types below
name	Symbol name (function/class/method/heading), if applicable
content	Raw source text
contextContent	Embedding input: content prefixed with breadcrumb, file imports, and (for methods) the class header
startLine / endLine	1-based line range in the original file
metadata	Extra data, e.g. { className: "UserService", breadcrumb: "// Class: UserService" } on method chunks

Chunk types

Type	Produced from
function	Function declarations, generator functions, arrow functions, Rust/Go/Python/Bash functions
class	Class declarations; also Rust struct and impl blocks, Ruby/Python module
method	Individual methods inside a class (Java, C#, PHP, Go receiver methods, …)
interface	TypeScript interface; also Rust trait, Swift protocol
type	TypeScript type alias; also Rust/Java/Kotlin enum
variable	Variable declarations
import	All import statements in a file, grouped as one chunk
section	Markdown heading section (H1–H3)
file	Whole-file fallback for small or structureless files (e.g. SQL)