---
title: API overview
description: What this reference covers, and how to run every request against your own vllm serve instance.
url: https://vllm-agent-docs.sudhanvasp.dev/api/introduction
---

# API overview

What this reference covers, and how to run every request against your own vllm serve instance.

> **Note:**
  Adapted from vLLM's [OpenAI-Compatible Server reference](https://docs.vllm.ai/en/latest/serving/online_serving/openai_compatible_server/) (Apache-2.0).

This reference documents the core HTTP endpoints exposed by `vllm serve` —
vLLM's OpenAI-compatible server. It's a scoped subset of vLLM's documented
API surface (chat completions, completions, embeddings, and models); vLLM
also exposes audio, scoring, tokenization, and realtime endpoints not
covered here. See [`openapi.yaml`](https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/openai/api_server.py)
in the vLLM repository for the complete, version-exact surface.

## Base URL

Every request in this reference runs against your own server, not a shared
demo — vLLM has no public hosted endpoint. Start one first (see
[Quickstart](/quickstart)):

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

```text
http://localhost:8000
```

## Try your first request

#### Start a local server

    Follow [Quickstart](/quickstart) if you haven't already — you need a
    running `vllm serve` process for the **Send** button below to reach
    anything.

#### Open an endpoint

    Pick **Create a chat completion** from the Chat group in the sidebar.

#### Send it

    Fill in `model` with the model you're serving (for example
    `Qwen/Qwen2.5-1.5B-Instruct`) and press **Send**.

#### Check auth, if you enabled it

    If you started the server with `--api-key`, add
    `Authorization: Bearer <key>`. Without `--api-key`, no auth is required.

## Regenerate this reference from a live server

This file was hand-authored from vLLM's documented endpoint set, scoped to
what [Serve an OpenAI-compatible endpoint](/guides/serve-openai-endpoint)
covers. For the complete, exact spec of your installed vLLM version, pull it
live from a running server and replace `openapi.yaml` at the repository
root:

```bash
vllm serve Qwen/Qwen2.5-1.5B-Instruct &
curl http://localhost:8000/openapi.json -o openapi.yaml
```

The endpoint pages, navigation, and playground regenerate from that file on
the next publish.