The Data-Centric Framework for Building Enterprise RAG Pipelines
Observe, Structure, and Assemble your Knowledge Graph.
Sayou Fabric is a framework for Deterministic Knowledge Graph Construction.
We focus on transforming unstructured data into structured assets by analyzing their intrinsic topology—syntax trees, file hierarchies, and metadata. This approach ensures that the constructed Knowledge Graph is precise, reproducible, and structurally sound, strictly following the data's native organization.
We prioritize the native structure of data to build a solid skeleton before processing the content.
- Code: Parsed via AST to deterministically link Classes, Methods, and Import dependencies.
- Media: Structured via Temporal Timeline and Semantic Segmentation.
- Docs: Hierarchically organized by Headers, Tables, and Layout Coordinates.
Inspired by robust enterprise architectures, every component adheres to a strict hierarchy to ensure stability at scale:
- Tier 1 (Interface): The immutable contract defining behavior.
- Tier 2 (Template): Standardized logic for ETL pipelines (Battery-included).
- Tier 3 (Plugin): Vendor-specific implementations (e.g., YouTube, GitHub, Notion).
graph TD
%% 노드 정의
Brain[Sayou Brain
< Orchestrator >]:::brain
subgraph Libs [Data Libraries]
direction LR
Conn[Connector]
Doc[Document]
Ref[Refinery]
Chunk[Chunking]
Wrap[Wrapper]
Assem[Assembler]
end
Core[Sayou Core
< Ontology & Schema >]:::core
%% 연결
Brain --> Libs
Libs --> Core
%% 정렬 힌트 (Libs 내부)
Conn ~~~ Doc ~~~ Ref ~~~ Chunk ~~~ Wrap ~~~ Assem
%% 스타일 적용
class Conn,Doc,Ref,Chunk,Wrap,Assem modules
Sayou Fabric consists of independent, loosely coupled libraries that work together seamlessly.
| Package | Version | Description |
|---|---|---|
sayou-core |
 |
Foundational layer defining Schemas, Ontology, and base components. |
sayou-brain |
 |
Manages the entire lifecycle of data processing pipelines. |
sayou-connector |
 |
Universal ingestor with auto-source detection and streaming support. |
sayou-document |
 |
Parses documents while preserving layout, styles, and spatial coordinates. |
sayou-refinery |
 |
Normalizes data formats and removes noise or PII. |
sayou-chunking |
 |
Splits text intelligently based on context (Code, Markdown, Time). |
sayou-wrapper |
 |
Maps raw data to the Sayou Ontology and creates standardized Nodes. |
sayou-assembler |
 |
Constructs semantic relationships (Edges) between Nodes. |
sayou-loader |
 |
Exports the constructed Knowledge Graph to Databases or Files. |
sayou-extractor |
 |
Retrieves context using Hybrid Search (Vector + Graph). |
sayou-llm |
 |
A unified adapter interface for various LLM providers. |
sayou-visualizer |
 |
Visualizes the pipeline flow and renders 3D Knowledge Graphs. |
You can install the entire suite via the orchestrator package:
pip install sayou-brainOr install individual components as needed for a lightweight setup:
pip install sayou-chunking sayou-documentpip install sayou-visualizerThe sayou-brain package provides high-level facades that abstract away the complexity of underlying modules. Choose the pipeline that fits your data source.
Use StandardPipeline for layout-preserving document analysis.
flowchart LR
%% 노드 스타일
classDef input fill:#fff3e0,stroke:#e65100;
classDef process fill:#f3e5f5,stroke:#7b1fa2;
classDef output fill:#e8f5e9,stroke:#2e7d32;
Input[PDF / Office]:::input
subgraph SP [Standard Pipeline]
direction LR
C(Connector) --> D(Document Layout Parse)
D --> R(Refinery)
R --> CH(Chunking)
CH --> W(Wrapper)
W --> A(Assembler)
end
KG[("Knowledge Graph")]:::output
Input --> C
A --> L(Loader)
L --> KG
%% 강조
style D stroke-width:3px,stroke:#d32f2f
from sayou.brain import StandardPipeline
result = StandardPipeline().process(
source="./reports/financial_q1.pdf",
destination="knowledge_graph.json",
)
print(f"Ingestion Complete. Processed: {result['processed']}")Use NormalPipeline for logic-based extraction from Video, Code repositories, or Web sources.
flowchart LR
%% 노드 스타일
classDef input fill:#fff3e0,stroke:#e65100;
classDef process fill:#e1f5fe,stroke:#0277bd;
classDef output fill:#e8f5e9,stroke:#2e7d32;
Input[YouTube / Code]:::input
subgraph NP [Normal Pipeline]
direction LR
C(Connector) --> R(Refinery)
R --> CH(Chunking)
CH --> W(Wrapper)
W --> A(Assembler)
end
KG[("Knowledge Graph")]:::output
Input --> C
A --> L(Loader)
L --> KG
%% 강조
style CH stroke-width:3px,stroke:#0288d1
from sayou.brain import NormalPipeline
result = NormalPipeline().process(
source="youtube://YOUTUBE_VIDEO_ID",
destination="./output/graph_data.json"
)
print(f"Graph Construction Complete. Nodes: {len(result['nodes'])}")The output is a structured JSON strictly following the Sayou Ontology, ready for Graph Databases or Vector Stores.
Click to expand JSON example
{
"nodes": [
{
"node_id": "sayou:doc:1_h_0",
"node_class": "sayou:Topic",
"attributes": { "schema:text": "Financial Summary Q1" },
"relationships": {}
},
{ "……" }
],
"edges": [
{ "……" }
]
}Observability is crucial for data engineering. Sayou Fabric provides built-in rendering tools to inspect your data topology and pipeline flow.
Monitor the execution flow. Visualize which plugins are activated and verify how data transforms across the pipeline.
Transform raw data into an interactive KG city.
- Analytics Knowledge View: Inspect strict node properties and relationships for debugging.
- Showcase Knowledge View: Visualize the cluster and semantic density of your knowledge base.
Sayou Fabric is currently in Public Beta (v0.4.0). We are actively stabilizing the core engine and expanding the adapter ecosystem to support more enterprise data sources.
Apache 2.0 License © 2026 Sayouzone
| Agent | Example | Description |
|---|---|---|
Code Diff AI Agent |
▶ | |
Code Analysis AI Agent |
▶ |