LlamaIndex Core

0.14.20 · active · verified Thu Apr 09

LlamaIndex Core provides the foundational interface and components for building LLM-powered applications, enabling users to connect large language models with their private or domain-specific data. It includes data structures, indexing tools, query engines, and basic abstractions for LLMs and embedding models. The current version is 0.14.20, with frequent, often daily, releases across its modular ecosystem.

Warnings

breaking Major architectural shift to a modular package structure in versions ~0.10.x and onwards. Core functionalities moved to `llama-index-core`, and all LLM, embedding, vector store, etc., integrations became separate packages (e.g., `llama-index-llms-openai`, `llama-index-embeddings-openai`).
Fix: Migrate imports and installations to use `llama-index-core` for base classes and explicitly install `llama-index-<component>-<integration_name>` packages for specific integrations. Update import paths from `llama_index.<component>.<integration>` to `llama_index.<component>.<integration>` (e.g., `from llama_index.llms.openai import OpenAI`).
breaking The `ServiceContext` class was deprecated and largely replaced by the global `Settings` object for configuration. While `ServiceContext` might still exist in some forms, `Settings` is the recommended way to configure LLMs, embedding models, chunk sizes, etc.
Fix: Replace `ServiceContext.from_defaults(...)` with direct assignments to `Settings.llm`, `Settings.embed_model`, `Settings.chunk_size`, etc. Explicitly pass components where granular control is needed.
deprecated Support for Python 3.9 has been officially deprecated and removed.
Fix: Upgrade your Python environment to 3.10 or higher. The library currently targets `<4.0,>=3.10`.
gotcha Many LlamaIndex components (LLMs, embeddings, vector stores, data loaders) default to using `openai` if not explicitly configured. This often leads to `openai` being a de-facto dependency for basic usage, requiring an API key even if not intended.
Fix: Always explicitly configure your desired LLM and embedding model via `Settings.llm` and `Settings.embed_model` or by passing them directly to constructors. Ensure relevant integration packages are installed (e.g., `llama-index-llms-anthropic`, `llama-index-embeddings-huggingface`).
gotcha When migrating from older versions, `Document` and `Node` structures might have subtle differences in metadata handling and content fields. For instance, `text` vs `content` or `extra_info` vs `metadata`.
Fix: Consult the migration guides in the official LlamaIndex documentation when upgrading across major architectural changes. Pay close attention to how document content and metadata are accessed and stored.

Install

pip install llama-index-core Install core package
pip install 'llama-index-llms-openai' 'llama-index-embeddings-openai' Install common integrations (e.g., OpenAI)

Imports

VectorStoreIndex

from llama_index.core import VectorStoreIndex

SimpleDirectoryReader

from llama_index.core.readers import SimpleDirectoryReader

Settings
```
from llama_index.core import Settings
```
ServiceContext was deprecated and largely replaced by the global Settings object or explicit passing of components.

OpenAI

from llama_index.llms.openai import OpenAI

OpenAIEmbedding

from llama_index.embeddings.openai import OpenAIEmbedding

Quickstart

This quickstart demonstrates loading data, configuring the LLM and embedding model via global `Settings`, creating a vector store index, and performing a simple query. Ensure you have the `OPENAI_API_KEY` environment variable set and the `llama-index-llms-openai` and `llama-index-embeddings-openai` packages installed.

import os
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings
from llama_index.llms.openai import OpenAI
from llama_index.embeddings.openai import OpenAIEmbedding

# Ensure you have your OpenAI API key set as an environment variable
# os.environ["OPENAI_API_KEY"] = "sk-..."
OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY', '')

if not OPENAI_API_KEY:
    raise ValueError("OPENAI_API_KEY environment variable not set.")

# Create a dummy data directory and file
if not os.path.exists("data"):
    os.makedirs("data")
with open("data/hello.txt", "w") as f:
    f.write("The quick brown fox jumps over the lazy dog.\n")
    f.write("LlamaIndex is a data framework for LLM applications.")

# 1. Load data
documents = SimpleDirectoryReader("data").load_data()

# 2. Configure global settings (LLM and Embedding Model)
Settings.llm = OpenAI(api_key=OPENAI_API_KEY, model="gpt-3.5-turbo")
Settings.embed_model = OpenAIEmbedding(api_key=OPENAI_API_KEY, model="text-embedding-ada-002")

# 3. Create an index
index = VectorStoreIndex.from_documents(documents)

# 4. Create a query engine
query_engine = index.as_query_engine()

# 5. Query the index
response = query_engine.query("What is LlamaIndex?")
print(response.response)

view raw JSON →