What it does create-llama is a CLI tool that generates a working LlamaIndex application from a few prompts. Pick a use case (Agentic RAG, Data Analysis, Report Generation), choose Next.js or Python FastAPI, and it spits out a pre-wired project with a shadcn/ui chat frontend, LlamaIndex Server backend, and a data/ folder ready for your files.

The interesting bit The dual-stack approach is unusual: the same CLI can scaffold either a TypeScript/Vercel stack or a Python/FastAPI stack, with the Python backend notably ingesting video and audio files that the TS version can’t yet handle. It’s essentially create-next-app translated to the RAG era — opinionated glue that saves you from wiring OpenAI embeddings to a chat interface by hand.

Key highlights

• Ships with pre-built use cases: Agentic RAG, Data Analysis, Report Generation
• Frontend uses shadcn/ui components in a chat-interface layout
• Next.js option deploys to Vercel in “a few clicks”; Python option uses FastAPI
• Data ingestion: drop files in data/, run npm run generate (TS) or uv run generate (Python)
• Defaults to OpenAI’s gpt-4.1 and text-embedding-3-large, but supports “dozens of other LLMs” via manual config edits
• Non-interactive mode available with CLI flags for scripting

Caveats

• Switching from OpenAI defaults requires manually editing settings.ts or settings.py; the --ask-models flag exists but the README doesn’t fully clarify how it interacts with manual edits
• Adding new data files requires re-running the generate command and restarting — no live reloading mentioned
• LlamaCloud integration is offered but the API key prompt suggests it’s optional; unclear what breaks if you skip it

Verdict Worth a spin if you want to prototype a LlamaIndex app without drowning in boilerplate. Skip it if you already have strong opinions about your stack — this tool is deliberately opinionated, and you’ll spend time unwrapping its choices if they don’t match yours.