Document Processing

The MastraDocument class handles document chunking and metadata extraction.

Constructor

text:

string

Document text content

metadata?:

Record<string, any>

Optional metadata about the document

Methods

chunk()

Splits document into chunks and optionally extracts metadata.

strategy:

'sentence' | 'paragraph' | 'fixed'

Chunking strategy to use

parseMarkdown?:

boolean

Whether to parse markdown syntax

metadataExtraction?:

object

Metadata extraction options (requires OpenAI)

boolean | TitleExtractorsArgs

boolean | SummaryExtractArgs

boolean | QuestionAnswerExtractArgs

boolean | KeywordExtractArgs

Response Types

The chunk method returns an array of document nodes:

interface DocumentNode {
  text: string;
  metadata: Record<string, any>;
  embedding?: number[];
}

Error Handling

try {
  const chunks = await doc.chunk({
    strategy: "sentence",
  });
} catch (error) {
  if (error instanceof DocumentProcessingError) {
    console.log(error.code); // 'invalid_strategy' | 'extraction_failed' etc
    console.log(error.details); // Additional error context
  }
}

Transition Pgstore