Headquartered in New York and Paris, Hugging Face is an open-source provider of machine learning technologies, allowing users to train, deploy, and share AI models using Hugging Face open-source libraries and the Hub.

**By partnering with Meilisearch Cloud, Hugging Face actively fulfills its visionary commitment to advancing the democratization of AI.**

![](https://unable-actionable-car.media.strapiapp.com/ezgif_com_video_to_gif_1_e697b647e1.gif)

> _“Today, Meilisearch is used by Hugging Face to power the discoverability of 300,000+ AI models, datasets, and demos. It’s really important for AI democratization because if the knowledge is out there, but you can’t access it - what’s the point?” -_ Mishig Davaadorj, software engineer at Hugging Face

Challenge
---------

On its platform, Hugging Face enables users to upload their AI models, datasets, and demos. With a vibrant community of users constantly engaging with the [Hugging Face Hub](http://hf.co) - a platform for sharing and searching for Machine Learning artifacts - the question of discoverability takes center stage.

The Hugging Face Hub hosts over 220,000 AI [models](https://huggingface.co/models) catering to a variety of [machine-learning tasks](https://huggingface.co/tasks), all neatly stored in repositories. These model repositories are designed to make the exploration and utilization of models as seamless as possible. Each AI model in the Hugging Face repository is accompanied by a model card, a project file containing valuable metadata, which plays a crucial role in enhancing discoverability, reproducibility, and sharing. Model cards also provide important information about AI models’ biases and limitations, model descriptions, and training guidelines, and serve as comprehensive guides for users seeking models on the Hub or uploading their own.

Before the introduction of Meilsearch, Hugging Face relied on a simple filtering and keyword search solution. However, it became evident that there was a growing demand for a more flexible and typo-tolerant full-text search solution. This solution needed to ensure higher default relevancy for every search query and leverage the additional attributes and metadata stored within model cards.

Why Hugging Face chose Meilisearch Cloud
----------------------------------------

Before integrating Meilisearch into its ML model repository, Hugging Face had already been utilizing Meilisearch's free-of-charge open-source solution as a search engine for its [open-source libraries](https://huggingface.co/docs)’ documentation, including transformers & diffusers. This documentation comprised approximately 500 pages and had been in use for over a year. Because the team had already gained positive experiences from implementing and operating Meilisearch, when the need for enhanced discoverability in model card searches arose, there was no requirement for additional testing or proof of concept.  

**1\. Customizability of the ranking rules**

During the evaluation process, the Hugging Face team also considered Mongo Atlas Search. However, they weren't satisfied with the customizations available for ranking. Several factors were taken into account in order to enhance model card discoverability, such as project names, descriptions, and the number of likes or downloads on each card. Meilisearch has demonstrated greater flexibility and adaptability in accommodating these search criteria.

**2\. Transitioning to Meilisearh Cloud for its ease of use**

Having achieved successful implementation of Meilisearch in its documentation, the Hugging Face team smoothly transitioned to incorporating Meilisearch Cloud into their model cards repository, benefiting from out-of-the-box relevancy.

**3\. Quality of support and infrastructure outsourcing**

The Hugging Face team chose Meilisearch Cloud, in part due to its dedicated support capabilities. Given that achieving highly relevant search results within the repositories is of utmost importance to Hugging Face, the decision to opt for the Cloud version was further influenced by entrusting the infrastructure to the expertise of Meilisearch Cloud specialists. Outsourcing the infrastructure allows the Hugging Face team to enhance their development speed.

Implementation
--------------

With Meilisearch already implemented for Hugging Face documentation, the expansion of the search solution to other use cases was smooth, once the ranking and internal rules were in place.

Today, the Meilisearch engine powers the discovery of 220,000 model cards, 38,000 datasets, and 60,000 demos in the Hugging Face repository. The keyword filtering mechanism has already been implemented on the Hugging Face homepage for users who are already familiar with specific model names.

However, with the implementation of Meilisearch, users are given the option to perform full-text searches as well. This search functionality operates not only on the model names and IDs but also includes the entire content of model cards.

The additional button to “Try full-text search” was implemented in order to minimize changes in the search behavior, where users are free to choose the search experience that most fits their preference.

![](https://unable-actionable-car.media.strapiapp.com/ezgif_com_video_to_gif_bce9ae142e.gif)

Vision
------

With the ongoing evolution of AI, Hugging Face anticipates a shift in search behavior and the methods people use to find the projects they need. Hugging Face is looking forward to incorporating Semantic Search into its documentation and model card searches. Currently, the platform is actively testing Meilisearch [VectorDB](/blog/langchain-semantic-search-tutorial/ as one of the potential solutions.

Hugging Face also anticipates gaining more detailed usage insights, encompassing information about users' daily search patterns and the level of activity certain model card projects receive over time.

Hugging Face facilitates AI accessibility with Meilisearch

At Meilisearch, we have long maintained our commitment to simplifying and streamlining the search process by prioritizing intuitive, user-friendly, and human-like search experiences.

In today's search technology landscape, _[keyword search](https://www.meilisearch.com/docs/learn/what_is_meilisearch/overview?utm_campaign=vector-search&utm_source=blog&utm_medium=launch#features)_ continues to play a pivotal role, swiftly locating content via specific keywords. Leveraging advanced algorithms and robust indexing, it ensures users find relevant information across vast data volumes.

However, traditional keyword approaches have inherent limitations in understanding user intent and at times can be overly predictable and simplistic. This is why we are introducing Meilisearch's _[vector search](https://www.meilisearch.com/docs/learn/experimental/vector_search?utm_campaign=vector-search&utm_source=blog&utm_medium=launch),_ which aims to challenge and enhance the familiar keyword search paradigm.

Vector search for AI-enhanced future
------------------------------------

The market-wide introduction of _vector search_ marks the arrival of a new era for search technology, with countless versatile applications. In _e-commerce_, it offers product recommendations based on similarities and multi-modal search capabilities such as image, audio, and video search. The fusion of vector search with conversational AI technologies opens doors for intuitive Q&A applications. Adding user data like geolocation and search history further expands its potential.

New search technology enhances relevancy alongside traditional keywords with semantic understanding going beyond simple keyword matching. It is capable of comprehending intent, returning results like **\[gloves\]** from a **\[warm clothes\]** query. These applications rely on _vector embeddings_—a type of data representation with semantic information that enables AI to gain understanding and maintain long-term memory. Stored in _vector databases_, embeddings represent different data dimensions, essential for understanding patterns and relationships. AI-powered vector database then leverages this information when matching search queries to answers.

Democratizing search
--------------------

In a data-rich world, effective content discovery is not a luxury - it’s crucial. Meilisearch's vision balances ambition with accessibility, aiming to make cutting-edge search technology inclusive. Key Meilisearch partners, such as [Hugging Face](/blog/hugging-face-democratizes-ai/ are already actively testing Meilisearch’s new vector database and storage capabilities.

> **Quentin de Quelen, Meilisearch’s CEO and Co-founder, said:** _“Meilisearch's hybrid search will democratize advanced search technology, focusing on simplicity and open-source customizability.”_

Building the future of search with vector search
------------------------------------------------

This release brings semantic and hybrid search capabilities to enhance applications for image, video, and audio search, product recommendations for e-commerce, precise personalization, and intuitive Q&A interactions.

Here are a few examples of potential applications made possible with the introduction of vector search:

*   **Semantic understanding:** Enable your search engine to act as a human-like assistant, delivering natural, context-aware results in a more conversational manner
*   **Multiformat capabilities:** Cover images, audio, and video, ensuring versatility.
*   **Personalization**: Matches preferences for tailored results, with a deeper understanding of user behavior and preferences,
*   **Precise Recommendations:** Enhance the discovery experience by suggesting similar items that align with users' genuine interests.
*   **Improved accuracy:** Improve relevancy and search accuracy through a deep understanding of each search query's meaning

Search evolution culminates in a hybrid search
----------------------------------------------

Meilisearch's journey leads to _hybrid search_, which combines traditional full-text efficiency with the innovation of semantic search and the adaptability of vector database. The hybrid approach enables search solutions to cater to a broad spectrum of use cases driven by customer-generated data sources. By integrating models from AI solution providers like OpenAI, Cohere AI, or Hugging Face, Meilisearch empowers users to calibrate _vector embeddings_ and infuse their business logic into the search engine.

Getting started
---------------

<Callout>
  For more guidance on how to get started, explore our [step-by-step tutorial on building semantic search with LangChain](/blog/langchain-semantic-search-tutorial/?utm_campaign=vector-search&utm_source=blog&utm_medium=launch)
</Callout>
```json
from langchain.vectorstores import Meilisearch
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.document_loaders import JSONLoader

# Load documents
loader = JSONLoader(
    file_path="./movies.json",
    jq_schema=".[] | {id: .id, title: .title, overview: .overview}",
    text_content=False,
)
documents = loader.load()

# Index documents
embeddings = OpenAIEmbeddings()
vector_store = Meilisearch.from_documents(documents=documents, embedding=embeddings)

# Search
query = "superhero fighting villains in a city corrupted by crime"
results = vector_store.similarity_search(
    query=query,
    k=3,
)

```

If you want to use Meilisearch vector search capabilities without LangChain, choose your favorite language and check out the [relevant SDK documentation here](https://www.meilisearch.com/docs/learn/what_is_meilisearch/sdks?utm_campaign=vector-search&utm_source=blog&utm_medium=launch).

Want the quality of open-source without the headache of managing your servers and service? [Sign up for a free trial](https://meilisearch.com/cloud?utm_campaign=vector-search&utm_source=blog&utm_medium=launch) with Meilisearch Cloud today.

Introducing Meilisearch's vector search and vector database to navigate the future of search

Vector search enables efficient retrieval of objects sharing similar characteristics. This AI-powered search technique uses embedding vectors. These vectors are mathematical representations of objects generated by machine learning models (like LLMs). Starting with the 1.3 release, Meilisearch supports storing and searching vectors.

<Callout>
  Meilisearch v1.3 is out! Read the [release notes](/blog/v1-3-release/?utm_campaign=vector-search&utm_source=blog&utm_medium=announcement)
</Callout>

Powering the Future of Search
-----------------------------

Vector search might be the dawn of a new era for search. Its use cases are numerous. In [ecommerce](https://ecommerce.meilisearch.com/?utm_campaign=vector-search&utm_source=blog&utm_medium=announcement), it enables offering recommendations for similar products. It also allows building multi-modal search, like image or audio search. Building upon conversational AI technologies enables the creation of Q&A applications. Combining vector search with user-provided information like geolocation and search history can power even more contextual search experiences.

Vector search is also the foundation for [semantic search](https://en.wikipedia.org/wiki/Semantic_search), which aims to understand the query’s meaning. Conversely, traditional lexical search only matches keywords. With semantic search, a `warm clothes` query could give results like `gloves`, `coat`, and more results related to winter clothing.

Vector search unlocks a world of new capabilities for search. Take a look at what some users have already implemented:

*   [similar videos recommendations](https://github.com/adrienpoly/rubyvideo/pull/19) on a site listing Ruby conferences
*   a [documentation chatbot](https://blazy-chat.vercel.app/) proof of concept using [GPT3.5](https://platform.openai.com/docs/models/gpt-3-5) and [LangChain](https://github.com/hwchase17/langchain)

<Callout>
  Meilisearch is available as a [LangChain vector store](https://python.langchain.com/docs/integrations/vectorstores/meilisearch).
</Callout>

Getting Started with Meilisearch Vector Store
---------------------------------------------

  
Starting with v1.3, you can use Meilisearch as a vector store. Meilisearch allows you to store vector embeddings alongside your documents conveniently. You will need to create the vector embeddings using your third-party tool of choice ([Hugging Face](https://huggingface.co/), [OpenAI](https://openai.com/)).

First, spin up a Meilisearch instance. You can [install Meilisearch locally](https://www.meilisearch.com/docs/learn/getting_started/installation?utm_campaign=vector-search&utm_source=blog&utm_medium=announcement#local-installation) or create a [Meilisearch Cloud](https://www.meilisearch.com/cloud?utm_campaign=vector-search&utm_source=blog&utm_medium=announcement) account.

Then, enable the _vector store_ experimental feature:

```Bash
curl \
  -X PATCH 'http://localhost:7700/experimental-features/' \
  -H 'Content-Type: application/json'  \
  --data-binary '{
    "vectorStore": true
  }'
```
<Callout>
  This guide uses \`curl\` to make HTTP requests to communicate with Meilisearch. In practice, we recommend using our [SDKs](https://github.com/meilisearch/integration-guides) instead.
</Callout>

Meilisearch now accepts a `_vector` field in your documents. Use it to store the vector embeddings corresponding to your document in this field.

```Bash
curl -X POST -H 'content-type: application/json' \
     'localhost:7700/indexes/songs/documents' \
     --data-binary '[
         { "id": 0, "_vectors": [0, 0.8, -0.2], "title": "Across The Universe" },
         { "id": 1, "_vectors": [1, -0.2, 0], "title": "All Things Must Pass" },
         { "id": 2, "_vectors": [0.5, 3, 1], "title": "And Your Bird Can Sing" }
     ]'
```

After storing your vectorized documents in Meilisearch, you can query them using the `search` or `multi-search` route. To do this, you need to compute the vector of your query (using a third-party tool) and send it in the `vector` field.

```Bash
curl -X POST -H 'content-type: application/json' \
   'localhost:7700/indexes/songs/search' \
   --data-binary '{ "vector": [0, 1, 2] }'
```

When using vector search, returned documents include a `semanticScore` field:

```Bash
{
  "hits": [
    { "id": 0, "_vectors": [0, 0.8, -0.2], "title": "Across The Universe", "_semanticScore": 0.6754 },
    { "id": 1, "_vectors": [1, -0.2, 0], "title": "All Things Must Pass", "_semanticScore": 0.7546 },
    { "id": 2, "_vectors": [0.5, 3, 1], "title": "And Your Bird Can Sing", "_semanticScore": 0.78 }
  ],
  "query": "",
  "vector": [0, 1, 2],
  "processingTimeMs": 0,
  "limit": 20,
  "offset": 0,
  "estimatedTotalHits": 2
}

```
<Callout>
  This API is experimental. You can help us improve it by sharing your feedback in [this Github discussion](https://github.com/meilisearch/product/discussions/677?utm_campaign=vector-search&utm_source=blog&utm_medium=textlink).
</Callout>

A First Step Toward Semantic Search
-----------------------------------

Vector search is our first step towards semantic search. But our long-term goal is to provide hybrid search—combining the benefits of full-text and semantic search to provide the most relevant search experience. Clément Renault, founder and CTO of Meilisearch, shared his thoughts on Github about [exploring semantic search](https://github.com/meilisearch/meilisearch/issues/3838?utm_campaign=vector-search&utm_source=blog&utm_medium=textlink) — read for a founder’s perspective. We can’t wait to share more with you!

Drop your email below to learn more about our progress with AI-powered search. We’ll keep you posted on all updates regarding vector search & semantic search.

Get Meilisearch AI updates straight to your inbox 💌
----------------------------------------------------

\* indicates required

Email Address \*

Marketing Permissions

Please select all the ways you would like to hear from Meilisearch:

Email

You can unsubscribe at any time.

We use [Mailchimp](https://mailchimp.com/legal/terms) to send emails.

We’re excited to walk our first steps toward semantic search. We can’t wait to hear your thoughts on integrating Meilisearch as a vector store. You can give your feedback in [this Github discussion](https://github.com/meilisearch/product/discussions/677?utm_campaign=vector-search&utm_source=blog&utm_medium=textlink).

You can stay in the loop by subscribing to our [newsletter](https://meilisearch.us2.list-manage.com/subscribe?u=27870f7b71c908a8b359599fb&id=79582d828e&utm_campaign=vector-search&utm_source=blog&utm_medium=textlink). To learn more about Meilisearch's future and help shape it, look at our [roadmap](https://roadmap.meilisearch.com/?utm_campaign=vector-search&utm_source=blog&utm_medium=textlink) and participate in our [Product Discussions](https://github.com/meilisearch/product/discussions/?utm_campaign=vector-search&utm_source=blog&utm_medium=textlink).

For anything else, join our developers community on [Discord](https://discord.meilisearch.com/?utm_campaign=vector-search&utm_source=blog&utm_medium=announcement).

I’ll see you there.

Meilisearch latest news and company updates

Hugging Face facilitates AI accessibility with Meilisearch

Introducing Meilisearch's vector search and vector database to navigate the future of search

Vector storage is coming to Meilisearch to empower search through AI