Brave's Search API powers your favorite AI apps 🌐

CV Deep Dive

Today, we’re talking with Subu Sathyanarayana, VP, Search Engineering at Brave.

Brave’s Search API has quickly become an essential tool for AI developers, search engines, and businesses looking for an alternative to Google and Bing. As Microsoft increased Bing API pricing and Google remained largely inaccessible, Brave stepped in to provide a cost-effective, independent search API with real-time web data.

The API is now used by top AI companies for both LLM inference and training, with applications in RAG, AI-driven search features, and citation generation. Companies like Anthropic, Cursor, Cline, and Windsurf use it to power AI chat experiences, while traditional search engines like Kagi and Swisscows rely on Brave Search as a core provider. The API has also been adopted in education and enterprise software, with use cases in platforms like Chegg and Turnitin.

In this conversation, Subu discusses how the Search API evolved into a key part of Brave’s ecosystem, how developers are using it today, and the opportunity in building an independent search engine at scale.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Subu 💬

Subu - welcome to Cerebral Valley! Brave’s Search API is being used by some of the biggest AI companies in the world. Talk us through the evolution of the Search API as a product within Brave’s wider org. 

To understand the evolution of the Brave Search API, it’s useful to revisit the history of Brave Search. What we now call Brave Search originated in 2008 at Cliqz, a company focused on developing a privacy-centric, European alternative to dominant search engines. Cliqz ceased operations in 2020, leading to the creation of Tailcat, which Brave acquired in 2021 to launch Brave Search. By 2023, Brave Search had become the default search engine in the Brave browser, now serving over 80 million users worldwide.

The Brave Search API emerged as a strategic extension of this platform. We saw an opportunity to offer businesses a viable alternative to Google and Bing, especially since these were the only widely known global search indices for developers in the West. Google’s API wasn’t directly accessible to most—developers had to rely on third-party scrapers—while Bing provided direct access but sharply increased API prices after Microsoft’s significant investment in OpenAI, making it less affordable for many.

Recognizing this gap, we built the Search API to provide an independent, high-quality, and cost-effective option for companies seeking to move beyond the Google-Bing duopoly. Today, the API is a cornerstone of Brave’s mission to deliver innovative, accessible search solutions across its ecosystem.

How would you describe the Search API to the uninitiated developer or AI team? 

The Brave Search API at its most basic level lets you write software that can search the Web. Calling the API is like going to your favorite search engine and entering a keyword. For regular search users, the results are formatted intuitively into a list of links and widgets that helps them find whatever information they’re looking for. The Brave Search API on the other hand is like a more raw and flexible state of that same data.

In AI, you’ll notice the leading trend and use case for chatbots is Web search. Perplexity, ChatGPT, Gemini, etc. — are all essentially combinations of traditional web search and LLMs. Speaking of LLMs, a search API is not just valuable for supplementing LLMs with information at the point of inference, they’re also highly useful for training. Is there any data set larger or more current than the billions of pages of the Web? The answer is no.

Who are the Search API’s most fervent users today? Who is finding the most value in using it on a regular basis? 

The Brave Search API is used by most of the top 10 AI companies, whether for training or at the time of inference. For some customers, their search features are powered exclusively by Brave Search.

When Anthropic announced Claude MCP, Brave Search was one of the first servers they demonstrated, and as a result Brave Search is used by tens of thousands of developers across more than a dozen apps such as Cursor, Cline, and Windsurf to assist with coding, as well as hundreds of developer projects and products using Claude. As of this March, the API has over 35,000 free and 2,700 paid customers subscribed to plans. 

Beyond pure AI businesses, the API is used by traditional search engines like Swisscows and Kagi, and in some less-obvious use cases like citations for education software with Chegg and Turnitin. Nearly everyone realized early on that training data can only go so far, and there’s a perennial need for real-time web data. 

Walk us through the Search API. What use-case should people experiment with first, and how easy is it for them to get started?

The easiest way to get started is by using low-code LLM orchestration apps like Dify.ai or FlowiseAI. With them, all you need is to input a Brave Search API key into a pre-built module to imbue the apps with search. For hobby users, the free plans will be more than enough. There are many non-developers using the Brave Search API in this way to create more personalized AI search experiences.

For more advanced developers that want to build projects at the scale of the web, our documentation can guide you through testing the API. I’d suggest playing around with basic RAG implementations to get a better sense of the inputs and outputs you’ll be building with. There are a variety of endpoints, query parameters, and request headers available to manage the quality and structure of your responses.

What are some use-cases for the Search API that illustrate its effectiveness? Any developer success stories your team would like to share?

One thing worth noting upfront: the first customer of the Search API is Brave Search itself (search.brave.com), which handles over 1.3 billion queries every month and serves millions of daily users. Developers can test it out themselves to see how effective it is and get a clear idea of what the API has to offer.

I’ve mentioned RAG applications a lot already, so it’s no surprise that this use case is the most common. Search and LLMs belong together kind of like peanut butter and jelly do. Our RAG customers choose us primarily because we’re able to support the massive scale that many of them require to deliver fast and appropriate web results as part of their chat experiences.

Secondary to chat bots, foundational LLM training has been a strength of our API as well. Our index of the Web is updated with millions of new pages everyday thanks in part to the Web Discovery Project, which is an opt-in feature in the Brave browser that anonymously reports popular pages as real people browse the Web. Real people seed the pages we include in the index, which means the quality reflects a more human experience. By avoiding a lot of the “made for bots” SEO junk on the Web, the Search API makes for a more human input into large training sequences.

There are a number of companies working on APIs for search. What sets Brave’s Search API apart from a product or technical perspective? 

While many companies offer search APIs, few provide independent, web-scale search indices of acceptable quality in the Western world—essentially just Google, Microsoft, and Brave. This narrows the choices to paying a premium, purchasing through a reseller, relying on grey-market scrapers, or opting for Brave’s Search API. What sets Brave apart is an API that is fast, high-quality, affordable, up-to-date, and independent, offering a compelling alternative to the competition.

How does the Search API actually work under the hood?

The Search API serves as an interface to Brave Search, a web-scale independent search engine. Its core functionality aligns with that of other web search engines and can be broken down into the following steps:

1. Crawling and Monitoring: Continuously monitor and crawl web pages to capture the latest changes across the internet.

2. Indexing: Construct multiple search indices using both keyword-based and vector-matching techniques to retrieve a candidate set of relevant pages.

3. Ranking and Filtering: Apply a machine learning (ML) model to rerank the candidate pages, prioritizing the best results while filtering out noise as effectively as possible.

4. Infrastructure: Maintain a robust infrastructure capable of handling hundreds of millions of daily queries while adhering to strict latency requirements.

For a deeper exploration of the challenges involved in building a search engine, we recommend checking out https://0x65.dev/, a blog by the engineers at Cliqz—the precursor to Brave Search. Though some content may not reflect the latest advancements in search technology, it remains a valuable resource for those interested in the subject.

The Search API provides access to most features available in Brave Search. Depending on the plan, it supports various result types—such as Autosuggest, Spellcheck, or Image Search—each delivering distinct outputs for the same query. Users can further customize results using a range of parameters, along with request and response headers that enhance output quality and provide additional useful information.

What has been the hardest technical challenge around building the Search API into the product it is today?

The hardest technical challenge was to get the search quality up to a standard acceptable to our users, many of whom were switching from either Google or Microsoft. Brave operates a search engine with a very small fraction of the resources these companies have, making it extremely challenging but ultimately rewarding to design and rethink the search problem.

User expectations continue to evolve, with long natural language queries seeking direct answers rather than references becoming increasingly common. The growing popularity of LLMs and RAG use cases present fresh challenges, requiring our team to adapt and enable better experiences. We continue to invest significant time and effort into our search quality, as it remains at the very core of what we do. We conduct extensive automated quality assessments, along with smaller-scale human evaluations using blind tests. These indicate that our search performance is competitive with Google and Bing, surpassing the latter in the US.

What makes having an independent global Search API so important to the Brave mission? 

Brave’s all about building a web that puts users first—think privacy, security, and the big one: choice. That’s where the Search API fits in. By giving businesses a solid option beyond Google and Microsoft, we’re shaking up the web so it’s not just a giant duopoly running the show. It is not just about throwing another search engine into the mix—we’re working closely with our partners to understand their specific needs and help them build cool solutions with our tech.

How do you see the Search API evolving over the next 6-12 months? Any specific developments that developers should be excited about? 

Brave Search and its API will evolve rapidly throughout 2025, with a focus on enhancing search quality and reducing latency to improve the developer experience. In the next 6-12 months, we’re prioritizing advancements such as better search relevance powered by advanced machine learning models, broader support for natural language queries, and tighter integration with technologies like large language models (LLMs) and retrieval-augmented generation (RAG). 

Developers can also expect an improved image search experience and better internationalization. At the heart of these updates is our commitment to providing an independent, affordable, and high-performing API that continues to stand out in the search landscape.

Anything else you'd like our readers to know about the Search API? 

I made it all this way without mentioning that the API is free to try with generous plans for business testing/hobby use. It’s also worth reiterating that we can and do work very closely with customers to tailor the API to their business needs— whether that be custom plans, features, or support. We’re always open to feedback on how the Search API can bring more value to our customers.

Conclusion

Stay up to date on the latest with Brave’s Search API, learn more about it here.

Read our past few Deep Dives below:

• Build specialized RAG agents with Contextual AI 🌐

• Remyx - Your AI Production Assistant 💡

• Qdrant's GPU-accelerated vector indexing is here 🔋🌐

• Athena is your AI-powered remote hire 🎛

• Doowii - your AI-first education platform 📖

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

All Events | Jobs