LlamaIndex Indexing

Index design determines whether retrieval finds useful evidence. Poorly prepared documents can look like a generation problem later.

#Documents and nodes

A Document is divided into Nodes. A node can carry text, file/page information and relationships. It is a unit for retrieval, filtering, citations and source tracing—not just a string fragment.

#Index choices

VectorStoreIndex embeds nodes for semantic retrieval. SummaryIndex keeps a sequence useful for document-wide synthesis. TreeIndex represents hierarchical organization. KeywordTableIndex supports term-oriented lookup. Check the installed version before choosing an integration.

#Persist and reload

pythonCopy
index.storage_context.persist(persist_dir="./storage")

from llama_index.core import StorageContext, load_index_from_storage
storage_context = StorageContext.from_defaults(persist_dir="./storage")
index = load_index_from_storage(storage_context)

#Connect a vector store

pythonCopy
import chromadb
from llama_index.core import StorageContext, VectorStoreIndex
from llama_index.vector_stores.chroma import ChromaVectorStore

db = chromadb.PersistentClient(path="./chroma_db")
collection = db.get_or_create_collection("my_data")
vector_store = ChromaVectorStore(chroma_collection=collection)
storage_context = StorageContext.from_defaults(vector_store=vector_store)
index = VectorStoreIndex.from_documents(
    documents, storage_context=storage_context
)

This fragment assumes documents, provider configuration and compatible integration packages already exist.

#Improve the foundation first

Small chunks can lose context; large chunks can introduce noise and cost. Treat defaults as a starting point. Add useful metadata such as section/source identifiers and evaluate representative queries before changing index types repeatedly.

#Next step

• Querying