LlamaIndex

What it does

LlamaIndex Framework is an MIT-licensed Python toolkit that takes in your documents and data sources (PDFs, databases, APIs, SQL) and turns them into indexes that an LLM can search when it answers. It produces question-answering systems with retrieved context, chatbots, structured extraction results, and tool-using agents, plus event-driven workflows that can be deployed as services.

Developers typically load files with a reader, build a vector store index, and query it in a few lines of code, or hand that index to an agent as a tool. The high-level API covers simple cases; lower-level modules (connectors, retrievers, query engines, rerankers) can be swapped or customised. It ships as a starter package or a lean core plus over 300 integration packages for LLM, embedding and vector store providers. Local models are supported, so a hosted API is optional.

It suits Python developers building retrieval or agent applications, from prototype to production. You install it with pip and supply your own model keys; the default index embeds with OpenAI unless you configure a local model. Built-in readers suit clean text, while scanned or complex documents may need the separate hosted LlamaParse service or the local LiteParse parser. The repository notes the company's main focus has shifted toward LlamaParse, though the framework remains available.

What it can do

• Connect to data sources and file formats such as APIs, PDFs, docs and SQL
• Build vector store indexes and query or chat engines over them
• Build agents that call tools, including RAG pipelines
• Orchestrate event-driven workflows with branching, retries and human-in-the-loop review
• Extract typed fields from text using Pydantic
• Run with local models instead of hosted APIs

Works with

• OpenAI
• Ollama
• Hugging Face embeddings
• Graphsignal

Example workflows

Question answering over a document folder

Input: a directory of files and an LLM/embedding key. Steps: load with SimpleDirectoryReader, build a VectorStoreIndex, create a query engine, ask a question. Output: an answer grounded in retrieved passages.

Fully local document chatbot

Input: documents plus a local Ollama model and Hugging Face embeddings. Steps: set Settings.llm and embed_model, build the index, query or persist to ./storage. Output: answers generated without a hosted API.

Before you choose

• Built-in readers suit clean text; scanned PDFs, forms and slide decks may need LlamaParse (hosted) or LiteParse (local parser)
• The default embedding model is OpenAI, so a key or local configuration is needed
• Index data is stored in memory by default unless you persist it

Listed under

• Data
• Data extraction
• Developers
• Document AI
• Document parsing
• Enterprise
• OCR

Get started

pip install llama-index, then follow the starter tutorial

1. Install with pip install llama-index, or llama-index-core plus chosen integration packages
2. Set a model provider key such as OPENAI_API_KEY, or configure local models via Settings
3. Follow the starter tutorial to build an index and an agent over your documents

The framework is MIT-licensed and free to use. Costs for LLM and embedding providers, vector stores and compute are separate. Hosted LlamaParse is a separate product and not priced here.

Source: developers.llamaindex.ai/python/framework, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• LlamaIndex Framework, by the makers of LlamaParse | Developer Documentation ↗developers.llamaindex.ai
• GitHub - run-llama/llama_index: LlamaIndex is the document processing platform for AI · GitHub ↗github.com

Company website

developers.llamaindex.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.