Skip to main content

​
Overview

SourceContent represents document chunks — uploaded documents split into searchable units. Each chunk maintains a reference to its original source, preserving context and traceability. For why structured memory matters over simple chunk search, see the concept overview.

​
Fields

FieldTypeDescription
idstringUnique identifier
contentstringChunk text content
sourceIdstringReference to original document
sourceTypestringDocument type (e.g., pdf, markdown, text)
metadataobjectAdditional metadata (page number, section, etc.)
createdAtstringCreation timestamp

​
Supported Formats

FormatExtensionsNotes
PDF.pdfPage-level chunking
Markdown.md, .mdxSection-level chunking
Text.txtParagraph-level chunking

​
Chunking

Documents are automatically split into chunks optimized for semantic search. Each chunk retains metadata linking it back to its position in the original document. No manual configuration is required — the chunking process is fully automatic.
Each chunk preserves its relationship to the original source, so you can always trace a search result back to the exact page, section, or paragraph it came from.

​
Relationship to Knowledge

Facts extracted from SourceContent can become Knowledge. The original SourceContent serves as the source reference, maintaining a clear provenance chain from raw document to confirmed fact. 
Document: "API Specification v2.pdf"
  |-- SourceContent chunk 1: "Authentication section"
  |     --> Knowledge: "API uses OAuth2 authentication"
  |-- SourceContent chunk 2: "Endpoints section"
  |     --> Knowledge: "Base URL is /api/v2/"
  |-- SourceContent chunk 3: "Error codes section"
        --> Knowledge: "Error 429 means rate limit exceeded"

Data Types

Concept overview of all data types

Knowledge API

SourceContent is managed alongside Knowledge