Agentic GraphRAG. Integrating Knowledge Graphs, Reasoning, and Agency for Enterprise AI Anthony Alcaraz, Sam Julien

Dostępność produktu

Produkt został zweryfikowany pod kątem ułatwień dostępu więcej przeczytasz w deklaracji dostępności.

Spis treści książki

• Preface
  • Who Should Read This Book
  • Why We Wrote This Book
  • Navigating This Book
  • Conventions Used in This Book
  • Using Code Examples
  • OReilly Online Learning
  • How to Contact Us
  • Acknowledgments
• I. Foundations of Graph-Based Agentic Systems
• 1. The Crisis of Enterprise Agentic AI
  • Defining Agentic AI
    • What Is an Agent, Anyway?
    • Classifying Agentic Systems: The Workflow-Agent Spectrum
    • The Agent Constraint Triangle
    • Agent Architecture
  • The Limitations of Vector-Based Retrieval
    • A Brief Overview of Vector Retrieval
    • Where Vector RAG Succeeds and Fails
      • Wont a larger context window solve this?
      • Consequences for vector-based agents
  • GraphRAG: The Evolution of Contextual Retrieval Systems
    • A Brief Overview of GraphRAG Architecture
    • Where GraphRAG Succeeds and Struggles
      • Context control
      • Flexibility
      • Better reasoning
      • Explainability and traceability
      • Self-evolution data flywheel
      • Where GraphRAG struggles
    • Hybrid GraphRAG
    • Learned Graph Reasoning Models
  • Summary
• 2. Agentic Graph Architecture Foundations
  • From Strings to Things
  • The Dual-Graph Architecture
    • The Vertical Knowledge Graph
    • The Horizontal Workflow Graph
      • Defining the harness
      • Splitting a workflow into nodes
    • Where the Two Graphs Meet
    • A Note on Context Graphs
  • The Eight Pillars of Agentic Graph Architecture
    • Setup: Building the Digital Twin of a Tech Stack (Part II)
      • Knowledge representation (Chapter 3)
      • Memory systems (Chapter 4)
    • Action: From Firefighting to Foresight (Part III)
      • Reasoning with graphs (Chapter 5)
      • Planning systems (Chapter 5)
      • Tool orchestration (Chapter 6)
      • Structured output engineering (Chapters 5 and 6)
    • Production: Creating a Self-Healing System (Part IV)
      • Self-evolution and evaluation (Chapter 7)
      • Optimization (Chapter 8)
  • The Autonomous DevOps Agent
    • The Initial State
    • The Transformation Arc
      • Part II: Building the digital twin
      • Part III: From firefighting to foresight
      • Part IV: Creating a self-healing system
    • A Real-World Parallel
    • What the Completed Agent Looks Like
  • Summary
• II. Setup: Building the Digital Twin of a Tech Stack
• 3. Graph-Based Knowledge Modeling for Agentic Systems
  • Knowledge Graph Foundations
    • Components of GraphRAG Systems
    • Building Versus Buying a GraphRAG System
      • Factors to consider
      • What remains
  • Graph Data Models
    • Types of Graph Data Models
    • Evaluating Graph Models
      • Formal reasoning
      • N-ary relations
      • Performance
      • Tool ecosystem
      • Constraint expressiveness
      • Putting it all together
      • Open world, closed world, and agent safety
      • Graph structures
  • Graph Design Principles for Agentic Systems
    • The Three-Graph Architecture for Agent Knowledge
      • Domain graph
      • Lexical graph
      • Subject graph
      • Why this architecture works
      • Why this architecture matters for agents
      • Multimodal knowledge graphs
    • Schema Design Patterns
      • Event-centric pattern
      • Contextual boundary pattern
      • Multiperspective pattern
      • Capability model pattern
    • Homoiconic Knowledge Representation
      • Metaknowledge structures
      • Executable knowledge patterns
  • Integrating with Existing Systems
    • Knowledge Organization and Ontology Fundamentals
      • The knowledge organization spectrum
      • Ontology core components
      • From access to context
    • Creating a Unified Semantic Foundation
      • Enterprise ontology challenges and practical solutions
      • Merging multiple ontologies
      • Annotating ontologies to control agent behavior
      • Upper ontologies
      • Iterative ontology creation with AI assistance
    • Entity Resolution: The Foundation of Agent Knowledge
      • Why traditional approaches fail
      • Evidence-based resolution versus generalization-based AI
      • Entity resolution as graph building blocks
      • Edge cases
  • Building the Knowledge Graph
    • Extraction Approaches for Heterogeneous Sources
      • Structured database integration
      • LLM-based extraction from unstructured text
      • LLM-based knowledge graph construction frameworks
        • The incremental approach: iText2KG
        • Document-level extraction: RAKG
        • Temporal dynamics: ATOM
    • Automating Knowledge Graph Construction with Multi-Agent Systems
    • Entity Resolution and Linking Across Graphs
      • The entity linking challenge
      • Three-stage linking pipeline
      • Quality metrics and edge cases
  • Knowledge Representation in Practice: DevOps Agent
    • Context Engineering for DevOps Automation
    • The DevOps Knowledge Representation Challenge
    • Designing the DevOps Ontology
      • Multi-cloud abstraction through canonical types
      • Applying schema patterns to infrastructure
    • Populating the Knowledge Graph: Building the Digital Twin
      • Structured data integration from cloud providers
      • Code and configuration integration
      • Observability platform integration
      • Operational metadata for business context
      • Integrating organizational ontologies and controlled vocabularies
    • Homoiconic Knowledge Representation for Agent Adaptability
    • Querying the Knowledge Graph: Making the Digital Twin Useful
      • Understanding service dependencies
      • Configuration drift detection
      • Impact analysis for infrastructure changes
    • Entity Resolution for Infrastructure Identity
    • Measuring Context Quality: Ensuring the Digital Twin Stays Accurate
  • Summary
• 4. Agentic Graph Memory Systems
  • Architecting Graph Memory Systems
    • The Problem: Why Your Current Memory Approach Fails
      • Sources of failure
      • Cache sharing in multi-agent systems
    • The Solution: Graph Memory
      • Graph memory in production
      • What graph memory gives you
      • Memory consistency models for agent coordination
  • Representing Memory as Graphs
    • Understanding the Architecture Evolution
      • The hierarchy of memory architecture
      • The computer architecture of agent memory
    • Essential Graph Components
      • Nodes: Your knowledge units
      • Edges: Your relationships
      • Subgraphs: Your contexts
    • Temporal Awareness: Making Memory Evolve
    • Implementation: Building Your First Graph Memory System
      • Initial integration
      • Handling complex relationships
  • Four Essential Memory Operations
    • Consolidation: From Experience to Knowledge
    • Indexing: Organizing for Speed
    • Updating: Handling Change Gracefully
    • Retrieval: Finding What Matters
  • Production Patterns and Architectures
    • The Letta (MemGPT) Approach: Hierarchical Memory
    • The A-Mem Pattern: Evolving Knowledge Networks
    • The Graphiti Pattern: Real-Time Performance
  • Comparing Production Systems
    • Cognee
    • Mem0
    • Zep
    • Choosing Your Approach
  • Production-Ready Features and Operations
    • Reasoning and Recommendations
      • Multi-hop reasoning
      • Intelligent recommendations
      • Debugging and maintenance
      • Health monitoring
      • Visualization and debugging tools
      • Conflict resolution
      • Performance optimization
    • Use Case Optimizations
  • Training Your Agents to Use Memory
    • The MEM1 Revelation: Let Them Learn, Not Follow Rules
    • From Static Rules to Learned Strategies
    • The Three Phases of Training
      • Phase 1: Task-oriented learning
      • Phase 2: Consolidation through constraints
      • Phase 3: Multi-objective mastery
    • Dynamic Memory Without Dynamic Schemas
    • Practical Training Strategies
      • Reward task success, not memory perfection
      • Use masked training for specialized behaviors
      • Implement curriculum learning
    • Evolving Your Memory System
      • Learned contextual updates
      • Discovered relationship patterns
      • Emergent consolidation
    • Moving Beyond Traditional Training
  • Adding Memory to the DevOps Agent
    • Designing the DevOps Memory Schema
    • Implementing Memory for Operational Context
    • Temporal Queries for Root Cause Analysis
    • Consolidation: Learning Patterns from Incidents
    • Training Memory Behavior with Reinforcement Learning
      • Defining the memory training environment
      • Configuring Nova Forge for memory training
      • What the agent learns
    • Production Considerations
  • Future-Proofing Your Implementation
    • Neurosymbolic Integration
    • Distributed Scaling
    • Continual Learning
  • Summary
• III. Action: From Firefighting to Foresight
• 5. Reasoning and Planning
  • Foundations: Why Graph-Based Reasoning Matters for Agent Development
    • The Two Graph Types: Horizontal and Vertical
      • Horizontal workflow graphs
      • Vertical knowledge graphs
      • The imaginary space
      • The pattern in the wild: Coding agents as dual-graph systems
    • Ontological Grounding: Keeping Your Agent in Reality
  • Retrieval Mechanisms
    • The Distracting Effect: Why Your Agents Get Misled
    • PathRAG: Structural Protection Against Distractions
    • R3-RAG: Teaching Your Agents When and How to Retrieve
    • GraphRAG Pattern Taxonomy
  • Reasoning and Generation
    • Interleaved Thinking: Real-Time Adaptive Reasoning
      • Where interleaved thinking helps and hurts
      • Implementation of interleaved thinking
    • Structured Generation: The Keystone of Reliable Communication
      • Why structured output changes everything
      • The compound power of structure and graphs
      • How Outlines transforms each component
      • Real-world impact
  • Pipeline Architectures
    • Sequential Pipeline: Linear Execution
    • Loop Pipeline: Iterative Refinement
    • Tree Pipeline: Parallel Exploration
    • Production Implementation Patterns
      • State management across pipelines
      • Error-handling strategies
      • Observability and debugging
    • Hybrid Architectures
      • Dynamic architecture selection
      • Nested pipeline composition
      • Graceful degradation
    • Event-Driven Orchestration: The Production Scaling Layer
  • Planning and Coordination
    • Why Planning Deserves Architectural Separation
    • Hierarchical Planning
    • Constraint-Guided Planning
    • Dynamic DAG Construction
    • Empirical Evidence: Graph Planning at Repository Scale
    • From Architecture to Daily Practice: Task Graphs at Developer Scale
    • The Multi-Agent Debate: Reconciling Parallel Execution
      • The real problem: Uncoordinated parallelism
      • When parallel execution makes sense
      • The architecture of controlled parallelism
  • DevOps Diagnostic Agent: Pulling It All Together
    • Building the Horizontal Workflow Graph
      • Workflow graph design
      • State-specific nodes and constrained context
    • Integrating the Knowledge Graph for Grounded Hypotheses
    • Constructing the Investigation DAG
    • A Production Incident Walkthrough
    • Why the Architecture Matters
  • Summary
• 6. Agentic Graph-Based Tool Orchestration
  • MCP: The Protocol That Changes the Game
    • Securing the Foundation: MCP Authentication
    • Agents as Tools: A Powerful Pattern
  • The Prompt Bloat Crisis: When More Tools Mean Less Intelligence
    • From Listing to Finding: The RAG Way
    • The Gateway Pattern
      • Enhanced tool representations: The Toolshed approach
      • Collaborative tool retrieval: COLT
    • Skills: The Judgment Layer
      • Skill quality evaluation
      • Integrating quality ratings into graph-based retrieval
      • Scale and retrieval
    • Choosing the Right Primitive: CLIs, MCPs, and Skills
      • The wrap-everything argument
      • Three primitives, three audiences
      • The personal-to-enterprise gradient
      • Per-agent access control
      • Convergence, not competition
  • The Knowledge Graph Necessity: Tools That Understand Each Other
    • Tools as Graph Nodes
    • Tool Dependencies and Type Matching
    • Securing Data Flow with Information Flow Control
  • Evolving Tool Ecosystems
    • Ontology-Driven Evolution: When Tools Define Themselves
      • Usage analytics
      • Pattern recognition
      • Automated improvement
    • The Trust Problem: When Tools Game the System
      • The solution: Verification-based trust
      • DRAFT: Learning what tools actually do
    • Metatooling: When Tools Create Tools
  • Orchestration at Scale
    • The Intelligent Orchestrator: Modernizing Classical SaaS
    • Hierarchical Orchestration: When Scale Demands Structure
    • Functional Clustering: Resilience Through Redundancy
    • The Enterprise AI OS Layer: Infrastructure for the Synthetic Workforce
  • Learning and Advanced Patterns
    • The Memory Layer: Learning from Every Interaction
    • Advanced Patterns: The Nested Execution Challenge
    • Performance at Scale: Making It Real
    • Context Governance: The Missing Layer in Tool Orchestration
      • The problem: Configuration drift at scale
      • Three solution architectures
      • Federated context architecture
      • Implications for tool orchestration
  • Tool Orchestration in Practice: The DevOps Agent
    • Registering Tools in the Knowledge Graph
    • Structured Output for API Integration
    • The Latency Investigation, Continued
  • Summary
• IV. Production: Creating a Self-Healing System
• 7. Self-Evolution and Evaluation
  • The Foundation of Self-Awareness
    • The Execution Graph: The Agents Autobiography
    • The Multi-Layered Evaluation Framework: A Cognitive Autopsy
      • Layer 0: The zero-shot hallucination gate
      • Layer 1: The context evaluator
      • Layer 2: The cognitive fault isolator
      • Layer 3: The tool-integrated reasoning judge
      • Nontraditional signals: What correctness metrics miss
      • A note on mechanistic interpretability
      • Proactive truthfulness detection: The dual-pathway model
    • A Taxonomy for Self-Evolution
      • What evolves
      • When evolution fires
      • How the agent learns
      • Where evolution applies
      • Connecting the axes
  • Targeted Self-Improvement
    • Choosing the Right Intervention: Prompt, Fine-Tune, or Constrain?
    • Semantic Backpropagation: Coherent Evolution Across the Graph
    • A Suite of Self-Improvement Frameworks
      • SEAL: Targeted data generation
      • Self-directed improvement: Moving from passive to active learning
      • Beyond prompts and weights: Code-level self-modification
      • TPT: Reinforcing what works
      • Reflect, retry, reward: Building metacognitive debugging skills
  • Safe Evolution in Production
    • The Full Evolutionary Loop
    • The RPO and Graduated Validation Protocol
    • Persistent Self-Learning Without Entropy Collapse
  • The DevOps Agent: Predictive Outage Detection in Practice
    • The Scenario: A Library Update Triggers a Prediction
    • Generating the Execution Graph
    • The Cognitive Autopsy
    • Semantic Backpropagation to the Causal Attribution Node
    • The Improvement and Its Effect
    • Inference-Time Knowledge Augmentation: Learning Without Retraining
    • Skills as Self-Improving Graph Objects
      • The add-cognify-search-learn pipeline
      • The five-stage self-improvement loop
      • Combined effect with XSkill
  • Common Pitfalls
  • Summary
• 8. Optimization
  • Selective Intelligence
    • The Case for Model Specialization
    • Routing Strategies
      • Static routing by node type
      • Threshold-based cascading
      • Learned routing
    • Measuring Cost-Performance Trade-Offs
  • Data Governance and Access Control
    • Subgraph-Level Access Control
    • PII and Data Retention
    • The Execution Graph as a Compliance Artifact
  • Production Systems and Maintenance
    • Schema Evolution
    • Node and Edge Lifecycle Management
    • Incremental Updates
    • Deployment Patterns
  • Hardware Acceleration
    • GPU-Accelerated Graph Operations
    • Inference Acceleration for Specialist Model Fleets
      • Parallelization
      • Optimized serving
      • Specialized hardware
      • Key-value cache as the binding concurrency constraint
    • Latency Budgets
  • The DevOps Agent: From Prototype to Production
    • Applying Selective Intelligence
    • Hardware Acceleration for Dependency Analysis
    • Governance for Infrastructure Data
    • The Completed Agent
  • Common Pitfalls
  • Summary
  • Conclusion
    • Where the Field Is Heading
    • Choosing the Right Architecture for Your System
    • Building What Comes Next
• Index

pokaż cały spis treści

ukryj recenzje