// the find
run-llama/create-llama
The easiest way to get started with LlamaIndex
A CLI scaffolder for LlamaIndex applications — run one command and get a working RAG or agentic app with a Next.js or FastAPI backend. It's the create-react-app of the LlamaIndex ecosystem, aimed at developers who want to skip the boilerplate and start experimenting with document Q&A or multi-step agent workflows.
The template coverage is genuinely broad — agentic RAG, deep research, financial report generation, form filling, and contract review are all real use cases with actual workflow code, not toy examples. Supporting both TypeScript and Python with their respective LlamaIndex libraries means you're not locked into one stack. The provider abstraction is clean: Anthropic, Gemini, Groq, Ollama, and others each get their own settings file, so swapping models is a one-file change. E2E tests using Playwright against generated apps catch regressions in the scaffolded output itself, which is the hard part of testing a generator.
It defaults to OpenAI and buries model configuration in generated files you have to manually edit — the '--ask-models' flag exists but isn't prominently surfaced, so most users will scaffold an OpenAI-dependent app before realizing they wanted something else. The generated code is a starting point that you immediately own: no upgrade path once you eject, so if LlamaIndex Server's API changes you're on your own. The Reflex UI templates (contract review, extractor) feel bolted on compared to the Next.js templates — different framework, different mental model, and the README barely explains why you'd choose it. At 1.5k stars it's lightly starred for something from the official LlamaIndex org, suggesting most users install it once and move on rather than treating it as a maintained dependency.