LlamaIndex
An AI app is only useful if it can see your information. LlamaIndex helps programmers feed documents, PDFs and databases to AI so it answers with your facts.
How you use it: A programmer installs the Python package and follows its documentation to connect documents and data.
In plain English
The README still describes LlamaIndex as a data framework for language-model apps. It offers connectors for sources such as APIs, PDFs, documents, and SQL, ways to index that data, and a query interface that returns an answer with the retrieved context attached.
The same README says the company's main focus has moved. LlamaParse is their platform for parsing and extraction, with a cloud signup. LiteParse is described as their effort at a free, fast text parser. The open-source framework remains available, and the note says you are welcome to use it.
You can install a starter package called llama-index, or llama-index-core plus only the integration packages you need. The README says there are over 300 integration packages for models, embeddings, and vector stores.
Practical ways to use it
Example 1
Ask questions against a folder of documents
The overview says the framework ingests PDFs, docs, and other formats, then answers a prompt with retrieved context.
Example 2
Connect a database you already have
The README lists SQL among the data connectors, next to APIs and documents.
Example 3
Parse difficult files in the hosted product
The README points at LlamaParse for document parsing of many formats, with a cloud signup for an API key. That product is separate from the open-source package.
Example 4
Pick only the integrations you need
The customized install path is llama-index-core plus a chosen provider package, such as the OpenAI example in the README.
Who it is for
Developer. Both install paths are Python packages. The hosted LlamaParse signup is a website, and the framework around it is still code.
Difficulty: Developer setup. A starter package exists. A useful app still means choosing a model, an embedding, and a place to store vectors.
Getting started
- 1
Read the focus note at the top of the README
The maintainers say their primary product energy is on LlamaParse and related parsers. The open-source framework is still published.
- 2
Install the starter package
The README names the PyPI package llama-index for a bundle of core plus a selection of integrations. Use llama-index-core if you want to add integrations yourself.
- 3
Follow the framework docs, not only this README
The README says it is updated less often than the documentation, and it links the developer docs for the Python framework.
- 4
Keep hosted parsing separate in your mind
LlamaParse and related cloud features have their own signup. You can use them with the framework or alone, and they have their own pricing.
What you need first
- Accounts
- A model provider key for most apps. LlamaParse needs the cloud account linked in the README if you want that parser.
- Hardware
- Python on a normal development machine for the framework. The README does not set a GPU requirement for the library itself.
- Cost
- The framework license on GitHub is MIT. LlamaParse is a cloud product. The README tells you to sign up and does not paste a price table.
Cautions
- The README says the company's focus has shifted toward LlamaParse. Do not assume every framework example is the path they are investing in most.
- The README itself warns that it lags the docs. When the two disagree, trust the docs link.
- A retrieval app can send private documents into a model prompt. Know which provider receives the text.
Questions people ask
What does LlamaIndex do?
It is a Python framework for connecting your documents and data to language-model apps, including search over what you ingested. The company also offers LlamaParse, a hosted document parser.
Who is LlamaIndex for?
Developers building apps that need private or company documents. It is a toolkit, not a desktop chat product.
What do I need first?
Python and the llama-index package, plus a model you can call. Read the note in the README about the shift toward LlamaParse.
What should I watch out for?
Use the official docs when they differ from the README. Remember that text you retrieve may be sent to the model provider you configured.