Flows (TrustGraph)

Flows in TrustGraph are processing pipelines that define how data moves through the system, from raw input to structured knowledge.

Flow Architecture

Input → Stage 1 → Stage 2 → Stage 3 → Output
        ↓         ↓         ↓
      Process   Transform  Store

Example Ingestion Flow

// Define a custom flow
const ingestionFlow = await trustgraph.flows.create({
  name: "document-to-graph",
  stages: [
    {
      name: "extract-text",
      processor: "pdf-extractor",
      config: {
        ocr: true,
        language: "en"
      }
    },
    {
      name: "extract-entities",
      processor: "ner-extractor",
      config: {
        model: "en_core_web_lg",
        entityTypes: ["PERSON", "ORG", "GPE"]
      }
    },
    {
      name: "extract-relationships",
      processor: "relation-extractor",
      config: {
        model: "relation-bert",
        minConfidence: 0.7
      }
    },
    {
      name: "build-graph",
      processor: "graph-builder",
      config: {
        deduplication: true,
        linkingStrategy: "semantic"
      }
    },
    {
      name: "index-vectors",
      processor: "vector-indexer",
      config: {
        embeddingModel: "text-embedding-3-large"
      }
    }
  ]
});

// Execute flow
await ingestionFlow.process({
  source: "documents/report.pdf"
});

Query Flow

// Query flows process user requests
const queryFlow = await trustgraph.flows.create({
  name: "graph-rag-query",
  stages: [
    {
      name: "vector-search",
      processor: "vector-retriever",
      config: { topK: 10 }
    },
    {
      name: "graph-traversal",
      processor: "graph-expander",
      config: { depth: 3 }
    },
    {
      name: "context-fusion",
      processor: "context-assembler",
      config: { strategy: "weighted" }
    },
    {
      name: "llm-generation",
      processor: "llm-generator",
      config: { model: "gpt-4-turbo" }
    }
  ]
});

// Execute query flow
const result = await queryFlow.process({
  query: "How does Company A affect Market B?"
});

Flow Classes

Flows can be grouped into classes for reusability (see Flow Classes).

Benefits

• Modularity: Reusable pipeline stages
• Flexibility: Easy to modify or extend
• Monitoring: Track performance of each stage
• Parallelization: Stages can run in parallel when possible

See Also

• Flow Classes
• Collections
• Knowledge Cores