---
title: Quickstart
description: Install vLLM and serve your first model with one command.
url: https://vllm-agent-docs.sudhanvasp.dev/quickstart
---

# Quickstart

Install vLLM and serve your first model with one command.

> **Note:**
  Adapted from vLLM's [Quickstart guide](https://docs.vllm.ai/en/latest/getting_started/quickstart.html) (Apache-2.0).

vLLM installs with `pip` (or `uv`, the faster alternative) and serves a model
from a single command. This page gets you from nothing to a running,
OpenAI-compatible endpoint.

## Before you begin

- A Linux machine with an NVIDIA GPU (vLLM also supports CPU, AMD, and other
  backends — see [supported models](/reference/supported-models) for backend notes)
- Python 3.9–3.12
- Enough disk space to download a model from Hugging Face

#### Install vLLM

    The recommended path uses [`uv`](https://docs.astral.sh/uv/), a fast Python
    environment manager:

    ```bash
    uv venv --python 3.12 --seed
    source .venv/bin/activate
    uv pip install vllm --torch-backend=auto
    ```

    Prefer conda? This works too:

    ```bash
    conda create -n myenv python=3.12 -y
    conda activate myenv
    pip install --upgrade uv
    uv pip install vllm --torch-backend=auto
    ```

#### Start the server

    Launch vLLM with a small model to confirm everything works:

    ```bash
    vllm serve Qwen/Qwen2.5-1.5B-Instruct
    ```

    The OpenAI-compatible server starts at `http://localhost:8000` by default.
    The first run downloads the model from Hugging Face, so it can take a few
    minutes depending on your connection.

#### Send a completion request

    In a second terminal, call the completions endpoint:

    ```bash
    curl http://localhost:8000/v1/completions \
        -H "Content-Type: application/json" \
        -d '{
            "model": "Qwen/Qwen2.5-1.5B-Instruct",
            "prompt": "San Francisco is a",
            "max_tokens": 7,
            "temperature": 0
        }'
    ```

#### Verify the result

    You should see a JSON response with a `choices[0].text` completion
    continuing the prompt. If the request hangs or errors, check
    [Troubleshooting](/troubleshooting) — the most common cause at this stage
    is the model still downloading, or insufficient GPU memory.

## Choose the next task

Now that the server runs, go deeper on the API it exposes in
[Serve an OpenAI-compatible endpoint](/guides/serve-openai-endpoint), which
covers the chat completions endpoint and both the curl and Python OpenAI
client paths.