run-llama /

LlamaIndex

Live map
View repository
Snapshotmain · afd0fef

Documents become indexed nodes; queries retrieve a relevant subset that an LLM synthesizes into an answer.

Click a building or route
Choose a path

Follow source text through transformations into vector storage.

System map7 components · 8 connections
130%
LlamaIndex: ingest once, retrieve at query time architectureUse the top lane for ingestion and storage, then follow the lower lane from a query through retrieval to synthesis.text, metadata, and do…documents or nodeschunks, embeddings, re…Data sourcesIngestionpipelineTransformationsVector indexRetrieverResponsesynthesizerQuery engine
Connections
BuildingsEntryServiceComputeStorageRuntimeToolingExternal

Terms, in plain English

NodeA document-derived unit of text or data with metadata and relationships.

How to study this repo

  1. 1
    Ingest one document

    Track its ID through splitting, embedding, and index insertion.

  2. 2
    Inspect retrieval

    Follow a QueryBundle into vector search and back as NodeWithScore values.

  3. 3
    Compare synthesis modes

    Read two synthesizers and note how each partitions context across LLM calls.