LlamaIndex 0.14.24 Install Guide [LangChain Alt]

The #1 mistake with a LangChain alternative install is treating llama-index like one global pip package. People install on the system interpreter, mix leftover LangChain pins, skip a virtualenv, then hit import errors or OpenAI auth failures and blame the framework.

LlamaIndex 0.14.24 (as of the Aug 19, 2026 PyPI/GitHub release) is a namespaced stack for RAG and document workflows. You want it when retrieval over your files is the hard part – not another alternatives essay. This is the deploy path that gets 0.14.24 running.

System requirements before you touch pip

Per PyPI, the package requires Python >=3.10 and <4.0. Older interpreters fail the resolver cold.

• OS: Windows 10+, macOS, or Linux with a supported Python
• CPU/RAM: fine for API-only RAG on a normal laptop; local embedding/LLM stacks need whatever those models demand
• Disk/network: enough space for the starter bundle plus whatever integrations you add; pip once, then API or local model traffic
• API keys: none required to install; default starter path expects OPENAI_API_KEY at query time

Official install docs also note that llama-index-core pre-bundles NLTK and tiktoken data so those pieces do not phone home at runtime. The framework stays light. Local Ollama or Hugging Face models are a separate hardware problem.

Official download sources for LlamaIndex 0.14.24

Stick to the project’s own channels:

• PyPI (recommended):https://pypi.org/project/llama-index/ – 0.14.24 (released Aug 19, 2026)
• GitHub:run-llama/llama_index – tag v0.14.24
• Docs:Installation and Setup

No binary installer. No single blessed “run LlamaIndex” server image as the primary path either – docs talk about packaging your app. Docker is DIY after pip works.

Install LlamaIndex 0.14.24 step by step

Clean project env. Non-negotiable.

# 1) Project folder + venv (Python 3.10+)
mkdir llamaindex-deploy && cd llamaindex-deploy
python3 -m venv .venv

# macOS/Linux
source .venv/bin/activate
# Windows PowerShell
# .venvScriptsActivate.ps1

# 2) Upgrade packaging tools
python -m pip install --upgrade pip setuptools wheel

# 3) Pin the starter bundle
pip install "llama-index==0.14.24"

Starter bundle, not a monolith: official install docs say that meta-package pulls llama-index-core, llama-index-llms-openai, llama-index-embeddings-openai, and llama-index-readers-file.

Want local models, not OpenAI?

pip install llama-index-core llama-index-readers-file 
 llama-index-llms-ollama llama-index-embeddings-huggingface

From source (contributors / bleeding edge):

git clone https://github.com/run-llama/llama_index.git
cd llama_index
# install Poetry first, then:
poetry self add poetry-plugin-shell # if shell plugin missing
poetry shell
pip install -e llama-index-core
# then editable installs for the integrations you need

First-time configuration (minimum viable)

Defaults still chat with OpenAI. Same shell as the venv:

# macOS/Linux
export OPENAI_API_KEY="sk-..."
# Windows CMD
set OPENAI_API_KEY=sk-...
# or load from .env in code

Smoke script (smoke_test.py) after you drop a .txt/.md/.pdf into ./data:

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader

documents = SimpleDirectoryReader("data").load_data()
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()
print(query_engine.query("What is this document about?"))

Pro tip: In production, pin the meta package and what you import (llama-index-core==..., LLM/embedding packages). The umbrella moves; locked pins stop surprise breaks on the next deploy.

Ollama path: install those integration packages, then set Settings.llm / Settings.embed_model. Starter OpenAI defaults will not apply.

Ever notice how “pip exited zero” and “the query returned an answer” are different jobs? That gap is where first-day frustration usually lives.

Verify the install works

pip show llama-index llama-index-core
python -c "from llama_index.core import VectorStoreIndex, SimpleDirectoryReader; print('core imports OK')"
python smoke_test.py

llama-index should read 0.14.24 (or your pin). smoke_test.py should print a real answer. Imports alone do not prove embeddings + LLM credentials.

Common install errors and fixes

1. No matching distribution / resolver rejects Python – Python <3.10. Install 3.10-3.12, recreate the venv.
2. ModuleNotFoundError: llama_index or missing readers – Wrong interpreter or partial install. which python, reactivate, pip install "llama-index==0.14.24" --force-reinstall --no-cache-dir.
3. Auth / connection errors on first query – Install succeeded; key missing or wrong. Docs still default to gpt-3.5-turbo + text-embedding-ada-002. Export OPENAI_API_KEY or switch integrations.
4. Conflicts on llama-index-embeddings-huggingface (torch pins) – Community threads report torch fights and heavy downloads. Fresh venv, install torch from the official index for your platform first, then the embedding package – or stay on OpenAI embeddings until you need local vectors.
5. Broken imports after an old upgrade (ServiceContext, pre-0.10 paths) – The v0.10 packaging split still haunts copied tutorials. New env, install 0.14.24 clean, imports under llama_index.core and integration namespaces (llama_index.llms.openai, etc.).

Upgrade and uninstall

pip install -U "llama-index==0.14.24"
pip check

Jumping from ancient 0.9.x code? Treat it as a migration: new venv, new install, fix imports. llamaindex-cli upgrade targeted the 0.10 cutover – do not assume it clears every later pin mess.

pip uninstall -y llama-index llama-index-core 
 llama-index-llms-openai llama-index-embeddings-openai 
 llama-index-readers-file
# also drop extra integrations you added
deactivate
rm -rf .venv # Windows: rmdir /s .venv

And delete ./storage if you persisted vectors while testing – leftover index folders are easy to forget.

The packaging split is why two version numbers can disagree in one env. Meta on PyPI is a thin starter; runtime truth sits in llama-index-core. Once you see that, a lot of “which package did I actually upgrade?” threads make sense.

FAQ

Is LlamaIndex a drop-in LangChain replacement?

No. Different APIs and defaults. Swap the dependency, then rewrite retrieval and agent code against LlamaIndex modules.

Should I install llama-index or only llama-index-core?

Use llama-index==0.14.24 for the starter OpenAI + file-reader bundle – quickest way to a working query engine. Go llama-index-core plus selective integrations when you refuse OpenAI defaults or want a thinner image (core + Ollama + Hugging Face embeddings in a locked Docker layer). Teams that also keep LangChain in the same project usually go custom so the graph stays readable.

Why does pip show llama-index look tiny compared to what got installed?

Because it is a meta/starter distribution. Weight lives in core and integrations. When you care what executes, check pip show llama-index-core. Tutorials that only pin the umbrella can still drift underneath.

Next: fresh venv → pip install "llama-index==0.14.24" → export OPENAI_API_KEY (or Ollama stack) → one file in data/ → run the smoke test until it prints an answer.