Frameworks and Orchestration · The LlamaIndex Ecosystem¶

Rather than focusing primarily on unified coordination of models and tools, LlamaIndex emphasizes how private data becomes high-quality context a model can use. It connects data sources, splitting, indexing, query engines, and event-driven Workflows into a data-centered orchestration pipeline.

This is not an exclusive division of responsibilities between frameworks. The current official Workflows documentation uses the independent workflows namespace and supports shared Context and typed state. Event-driven does not mean stateless, and serialization does not mean automatic recovery by default.

Chapters¶

How the pieces fit together¶

flowchart LR
    A[Data ingestion<br/>Document / Node] --> B[Indexes<br/>VectorStoreIndex and others]
    B --> C[Query engines<br/>Query Engine / Router]
    C --> D[Workflows<br/>Event + Context state]
    D -.Can be wrapped as a Tool.-> E[Agents in LangChain / other frameworks]

Reading suggestions¶

Start with Chapter 14 to understand how data becomes a retrievable structure, then read Chapter 15 to see how those structures support answers and multi-step processes.
To compare with LangChain, alternate between these chapters and their counterparts under docs/frameworks/01-langchain; both chapters point out differences in abstraction.
If you mainly want to understand state machines versus event-driven execution, start with Section 15.3, then consult the comparison table in Framework Selection and Portable Architectures.
To prepare for a technical discussion, try explaining how old nodes are removed after a document update, why a vector-store switch still requires regression testing, and how to prevent duplicate tool submissions after checkpoint recovery.

Return to AI Frameworks and Orchestration topics.