Web Browser Brave Adds Generative AI Assistant Leo
Brave has released a generative AI assistant named Leo for its desktop web browser. Leo’s chatbot interface leverages Meta’s Llama 2 and Code Llama large language models (LLMs) to summarize websites, generate text, and help with writing software code.
Leo employs the open-source Llama 2 model to answer queries, provide short synopses of websites, and compose new text. The decision to use Llama 2 and its code-centric derivative is notable as the Brave Summarizer introduced in March relied on an in-house LLM. With Leo, Brave wants to meld generative AI features with Brave’s existing privacy-focused browser. Brave engineered Leo to prevent tracking or profiling. Requests are anonymized, conversations are immediately discarded, and no login is needed. Brave began limited previews of Leo a few months ago, but the AI is now available by asking a question in the address bar and clicking “Ask Leo” or by clicking on the Brave Leo sidebar icon.
“It is increasingly common today to use AI to ask questions and get assistance. AI can be a powerful tool but it can also present growing concerns for data privacy and there’s a need for a privacy-first solution,” Brave CTO Brian Bondy explained. “Brave is committed to pairing AI with user privacy, and will provide our users with secure and personalized AI assistance where they already spend their time online.”
Leo & Claude
Leo is free for Brave users, but they can also upgrade by subscribing to Leo Premium for $15 a month. Leo Premium augments the AI with additional LLMs, including the larger and more powerful versions of Llama 2 and Code Llama, for faster and better responses. Anthropic’s Claude Instant LLM is the only other model included with Leo Premium for now, but Brave has plans to add others in the future. Claude Instant is faster and more powerful than Llama, 2, especially with regard to logical reasoning and math. Premium subscriptions utilize unlinkable tokens to validate access while maintaining privacy, keeping the browsing anonymous.
Brave claims Claude’s logical reasoning strengths combined with Brave’s privacy protections will help set a new standard for responsible, contextual AI. Leo Premium also comes with higher rate limits and priority access during peak usage. The companies also worked together on training Anthropic’s new Claude 2 LLM with Brave’s Search API as a way of making the AI better at retrieval augmented generation (RAG) queries.
“We’re excited that Brave is going to bring our state-of-the-art large language model Claude Instant directly into their browser. Integrating access to Claude seamlessly within Brave gives users assistance whenever they need it, right in the context of the page they are on,” Anthropic head of product research Matt Bell said. “Whether Brave users are analyzing text, editing documents, or writing code, they can now benefit from having personalized help from Claude at any time. This integration unlocks the potential for AI to enhance people’s browsing and productivity in an intuitive, contextual way.”
Brave is one of the rising number of web browsers with built-in generative AI chatbots. Opera has been one of the quickest to infuse LLM-powered features through its Aria generative AI assistant. Opera has been steadily adding new generative AI tools through Aria since its launch. Aria recently gained the ability to learn to mimic a user’s writing style based on submitted examples, which the privacy-centered Brave would be less likely to include. Of course, the biggest players are at the forefront, with Microsoft’s Bing AI chatbot and Google’s Bard AI at the top of the list.