How to use Hugging Face, in short: open the Spaces page to try AI apps in your browser, search the Models page when you want a specific model, and create an access token when you want to use one from code. The Hugging Face Hub hosts over 3 million models, 1 million datasets and 2 million AI apps, all open and publicly available, according to its documentation. (Hugging Face Hub documentation (opens in a new tab))

Hugging Face started in 2016 as a chatbot company and became known for Transformers, a code library it released publicly in 2018. It is often described as GitHub for AI: a place where people upload, download and share models instead of code. (Techopedia: What is Hugging Face (opens in a new tab))

For a normal user the site can feel like a library with no front desk. Most pages are written for developers, and many models need setup before they do anything. The good news is that you do not need to understand all of it. Three parts cover most of what people want: Spaces, model pages and the Playground.

The three things you find on the Hugging Face Hub

The Hub has three main kinds of content. Models are the AI models themselves, for text, images, audio and more. Datasets are collections of data for training and testing, with over 1 million public ones in more than 8,000 languages. Spaces are interactive apps that show a model working directly in your browser. (Hugging Face Hub documentation (opens in a new tab))

Every model page has a model card, which describes what the model does and its known limitations and biases. Some model pages also have an inference widget, a small box where you can type an input and see the result without leaving the page. (Hugging Face Hub documentation (opens in a new tab); Hugging Face docs: Inference Providers on the Hub (opens in a new tab))

How to try models on Hugging Face without writing code

1. Make a free account

Sign up at huggingface.co. You can browse without an account, but you need one to request access to gated models and to create tokens.

2. Start with Spaces

Open the Spaces page and look at the Space of the Week or search for a task, such as background removal or speech to text. Click a Space and use it like any web app. Spaces are built by the community, mostly with tools called Gradio and Streamlit, so quality and speed vary from one to the next.

3. Try a model on its page

Search the Models page for a model name or a task. If the model page has a widget, type into it to test the model. You can also filter the Models page by inference provider to see only models you can run this way.

4. Compare chat models in the Inference Playground

The Inference Playground at huggingface.co/playground lets you test chat models with your own prompts and compare them. It is the quickest way to see how an open model handles the kind of question you actually ask.

Widgets and the Playground both run on Inference Providers, a pay as you go service. According to the pricing page, free accounts get no included monthly credits and have to buy credits to use it, PRO accounts get $2.00 of credits a month, and Team and Enterprise organizations get $2.00 per seat. Credits are used before any pay as you go charges. (Hugging Face docs: Inference Providers on the Hub (opens in a new tab); Hugging Face docs: Inference Providers pricing (opens in a new tab))

How to get access to a gated model

Some popular models are gated: you must ask for access before you can download their files. You need to be logged in, and the request is made in the browser. Open the model page, fill in the form the authors ask for, and click the button to send the request, which by default reads Agree and send request to access repo. (Hugging Face docs: Gated models (opens in a new tab))

Some authors approve requests automatically and you get access at once. Others review them by hand, which can take time. If a request is rejected you cannot request that model again, and authors can remove access later. Some models also block users in EU countries based on IP address. (Hugging Face docs: Gated models (opens in a new tab))

How to use a Hugging Face model from code

5. Create an access token

Go to Settings, then Tokens, and create a token with inference permission. Store it as an environment variable named HF_TOKEN and never paste it into a public file.

Code 1: Chat with a model in Python

from huggingface_hub import InferenceClient
import os

client = InferenceClient(api_key=os.environ["HF_TOKEN"], provider="auto")

completion = client.chat.completions.create(
 model="[model id, for example deepseek-ai/DeepSeek-V3-0324]",
 messages=[{"role": "user", "content": "[your prompt]"}]
)
print(completion.choices[0].message.content)

The provider setting auto picks a provider for you. Hugging Face also offers an OpenAI compatible endpoint at router.huggingface.co/v1, so code written for the OpenAI client can switch to an open model by changing the base URL and key. That endpoint does not support image generation. (Hugging Face docs: Inference Providers on the Hub (opens in a new tab))

Code 2: Log in once from the terminal

hf auth login

Tip: Once you are logged in this way, downloads of gated models you have access to work in scripts too. You can also pass a token directly to most loading functions through their token parameter.

Tip: To run a model fully on your own computer instead, note that open models from the Hub also run in desktop apps such as LM Studio and Ollama. NeuraFeed has separate guides for both.

Where Hugging Face is weak for beginners: there is no single chat app like ChatGPT, Spaces vary a lot in quality and speed, and the best results usually need some code. Where it is strong: millions of open models sit in one place, and you can try many of them before you commit to anything. Start with Spaces, move to the Playground, and only write code when you know which model you want.