Introduction¶
Sayou Fabric is a framework for Deterministic Knowledge Graph Construction.
It transforms unstructured data into structured assets by analyzing their intrinsic topology — syntax trees, file hierarchies, and metadata. The result is a Knowledge Graph that is precise, reproducible, and structurally sound.
What Sayou Fabric Does¶
The full data-to-answer lifecycle is decomposed into discrete, composable stages:
| Stage | Package | Role |
|---|---|---|
| Ingestion | sayou-connector |
Fetches data from files, APIs, databases, SaaS |
| Parsing | sayou-document |
Converts documents to structured DOM with layout coordinates |
| Cleaning | sayou-refinery |
Normalizes, denoises, masks PII |
| Splitting | sayou-chunking |
Splits by native structure — AST for code, headers for docs |
| Mapping | sayou-wrapper |
Converts chunks to typed SayouNode objects |
| Assembly | sayou-assembler |
Resolves relationships between nodes into typed edges |
| Persistence | sayou-loader |
Writes the graph to VectorDB, Graph DB, or file |
The sayou-brain orchestrator wires these stages together. You invoke a Pipeline; Brain handles the rest.
Pipeline Modes¶
StandardPipeline — for documents where spatial layout matters (PDF, DOCX, PPTX).
Routes through sayou-document to preserve heading hierarchy, table boundaries, and page coordinates before chunking.
NormalPipeline — for code, video, and web content. Skips layout parsing. Routes directly to structural chunking — AST for code, temporal segmentation for video.
Scope¶
Sayou Fabric is a data layer framework. It produces a clean, queryable Knowledge Graph that any retrieval or generation layer can consume.