LlamaIndex is a Python data framework for building LLM and agentic applications over your own data. It addresses the gap between models pre-trained on public data and the private data an application needs, by supplying data connectors, index structures, and a retrieval and query interface that returns retrieved context alongside the model's output.
You build with a core package plus a chosen set of integration packages: the llama-index starter bundles core with a selection of integrations, while llama-index-core lets you add only the LlamaHub packages you need from a catalogue of more than 300 covering LLMs, embeddings, and vector stores. Imports are namespaced so that llamaindex.core.* refers to core modules and llamaindex. . refers to an integration.
A high-level API ingests and queries a directory of documents in a few lines; lower-level APIs let you customise or replace any module, including connectors, indices, retrievers, query engines, and rerankers. Indexes are held in memory by default and can be persisted to disk and reloaded through a storage context.
The framework is a companion to LlamaParse, a separate hosted platform for document parsing, structured extraction, indexing, and deployed document agents, which can be used with the framework or on its own.
Features
- Data connectors: ingest existing sources and formats such as APIs, PDFs, documents, and SQL
- Index structures: organise ingested data as indices or graphs for use with an LLM
- Retrieval and query interface: send a prompt, get back retrieved context and a knowledge-augmented response
- Modular integrations: swap LLM, embedding, and vector-store providers through separate packages, including local models via Ollama and Hugging Face embeddings
- Persistence: save an index to disk and reload it from a storage context
- Application-framework integration: connect to outer frameworks such as LangChain, Flask, or Docker
- LlamaParse platform: optional hosted Parse, Extract, Index, Split, and Agents services for document workflows
- Build-asset verification: bundled nltk and tiktoken caches carry GitHub build-provenance attestations
Integrates with
Integrated by
Agent Memory
neo4j-labs/agent-memory gives agents three memory layers in one graph: conversation histo...
AgentOps
Python SDK for AI agent monitoring, LLM cost tracking, benchmarking, and more. Integrates with most LLMs and agent frameworks including CrewAI, Agno, OpenAI Agents SDK, Langchain, Autogen, AG2, and CamelAI.
Future AGI
OpenTelemetry-based tracing and evaluation instrumentation for voice-AI applications.
GPTCache
Semantic cache for LLMs. Fully integrated with LangChain and llama_index.
Alternatives
724 Office
724-office runs a full AI agent loop in 3,500 lines of pure Python with three dependencies. Three-layer memory, MCP inte
ContextGem
Ask an LLM which clause a finding came from and it writes you a citation that reads right and points nowhere. ContextGem
Instructor
Structured, validated LLM output defined as Pydantic models, with automatic retries, streaming, and multi-provider support
