Why it matters
- 160+ data connectors eliminate custom data ingestion code — connect to Notion, Slack, PDFs, GitHub, and databases with a few lines of code.
- Production-tested retrieval patterns include advanced techniques (hybrid search, HyDE, auto-merging, re-ranking) that outperform naive chunking + embedding approaches.
- LlamaParse provides structured PDF parsing that preserves tables, headers, and lists — critical for high-quality RAG over complex documents.
- Active enterprise ecosystem with LlamaCloud for managed deployment and professional support.
Key capabilities
- Data loaders: 160+ connectors via LlamaHub; PDF, Notion, Slack, GitHub, databases.
- Node parsers: Intelligent document chunking with multiple strategies.
- Vector indices: VectorStoreIndex compatible with 20+ vector databases.
- Query engines: RAG pipelines with retrieval, synthesis, and citation.
- Retrieval modes: Semantic, BM25, hybrid, MMR, and custom retrievers.
- Advanced RAG: HyDE, auto-merging, parent document retrieval, re-ranking.
- Agents: ReAct, OpenAI function calling, and custom agent loops.
- LlamaParse: Cloud PDF parsing service with table/structure extraction.
- Streaming: Real-time streaming query engines.
- TypeScript: LlamaIndex.TS for JavaScript/TypeScript applications.
Technical notes
- Language: Python (primary); TypeScript (LlamaIndex.TS)
- License: MIT
- Install:
pip install llama-index
- GitHub: github.com/run-llama/llama_index
- Stars: 32K+
- LLMs: OpenAI, Anthropic, Cohere, Hugging Face, local (Ollama, LM Studio)
- Vector stores: Pinecone, Weaviate, Chroma, Qdrant, pgvector, 20+ more
Usage example
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.llms.openai import OpenAI
# Load documents and build RAG index
documents = SimpleDirectoryReader("./data").load_data()
index = VectorStoreIndex.from_documents(documents)
# Query with natural language
query_engine = index.as_query_engine(
llm=OpenAI(model="gpt-4o"),
similarity_top_k=5
)
response = query_engine.query("What are the key findings from the research papers?")
print(response)
Ideal for
- Python developers building RAG applications who need production-quality retrieval with many data sources.
- Teams with complex document ingestion needs (PDFs with tables, mixed formats, many source systems).
- Enterprise RAG deployments that need advanced retrieval techniques (hybrid search, re-ranking) for high-quality answers.
Not ideal for
- TypeScript/JavaScript-first teams — use LlamaIndex.TS or Vercel AI SDK + custom retrieval.
- Simple single-file RAG with no complex retrieval needs — direct LLM SDK may be simpler.
- Agent-heavy workflows without significant retrieval — LangChain has a stronger agent and tool ecosystem.
See also
- LlamaIndex TS — TypeScript version for Node.js and Next.js applications.
- Haystack — Alternative Python NLP pipeline framework; similar RAG capabilities.
- Weaviate — Vector database with native LlamaIndex integration.