Available Distributions

Distribution	Use Case	Inference
`starter`	General purpose, prototyping, production	Ollama, OpenAI, vLLM, Bedrock, and more
`nvidia`	NVIDIA NeMo Microservices	NVIDIA NIM
Custom	Your own provider mix	Any supported provider

Starter (Recommended)

The starter distribution works for most use cases. It includes all providers and auto-enables them based on available environment variables:

uv run llama stack run starter

It supports local inference (Ollama), cloud providers (OpenAI, Bedrock, Azure, etc.), and everything in between. See the Starter Guide for details.

Optimized for NVIDIA NeMo Microservices. See the NVIDIA Guide.

A minimal distribution that forwards requests to a remote Llama Stack server. See the Passthrough Guide.

Some partners host Llama Stack endpoints that you can connect to directly. See Remote-Hosted Distributions for available options.

Build your own distribution when you need a specific provider mix. See Building Custom Distributions.