The browser is the ultimate interface for AI

20 years since the ubiquity of the internet, a limitless source of knowledge available to all, something still feels missing. Instead of fostering enlightenment, it often feels like we're chasing fleeting information, our online experiences fragmented across a chaotic landscape of bookmarks, notes, and archived articles. This scattered data represents a wealth of untapped potential, waiting to be harnessed and transformed into meaningful insights.

The failure to capitalize on this potential isn't due to a lack of ambition but rather a shortcoming of our current tools. While web browsers have evolved in complexity, they remain fundamentally unchanged from their 1990s predecessors – essentially serving as digital bulletin boards that passively display information. The most widely used browsers are optimized for clicks and engagement, prioritizing advertising revenue over facilitating meaningful exploration and discovery. The times when exploring the digital vastness seemed magical are gone. It feels commercialized and spammy, and we often find themselves overwhelmed by the sheer volume of information, struggling to navigate the deluge of content effectively.

Moreover, these browsers track our every move, yet ironically never use all this data to actually help us. They fail to adapt to our individual needs, interests, and learning styles, making it challenging to build upon our existing knowledge and form meaningful connections between concepts. The most popular ones are specifically built to over-optimized driving clicks and engagement, and collect massive amounts of data for advertisers.

We've all felt the limitations:

• A Blank Slate: Every search feels like starting over, even when revisiting familiar territory. Browsers have no memory of our past explorations, no understanding of the connections we've made.
• The Information Overload: We're drowning in a sea of information, yet struggle to find the knowledge we seek. Search engines prioritize clicks over comprehension, trapping us in a cycle of shallow engagement and fleeting relevance.
• The Fragmentation of Thought: Our digital lives are scattered - bookmarks gather dust, notes languish in forgotten folders, brilliant ideas vanish into the ether of browser history.

I wanted something that can continuously help me think, grow and build upon what I'm actively doing and learning from it. In a more natural conversational-like way, almost like you would while co-working with a very smart colleague.

Portal is building a way to let your browser truly get to know you.

portal can recommend things personalized to you as it gets to know you over time

An Opportunity

For years, we've chased the dream of the "ultimate personal assistant". Sure, we had apps that could schedule meetings and set reminders, but true assistance — the kind that anticipates your needs, solves truly complex problems, and unlocks your full potential —always remained a just a leap away

We are constantly refining our tools to better align with our individual needs, preferences, and thought processes. The recent development of powerful Large Language Models represents a significant milestone in the evolution of human-computer interaction. They are designed to understand and generate human language with unprecedented accuracy and contextual awareness, enabling our tools themselves to communicate with us in a more natural and intuitive manner.

The capabilities of these models are particularly well-suited to addressing the challenges posed by the unstructured nature of the web as they are experts at converting all sorts of data into a unified schema - your personal context. This breakthrough unlocks a new era of personal assistance, enabling a digital partner that can:

• Learn your work style, anticipating needs before you even realize them: By observing your browsing habits, communication patterns, and the types of content you engage with, an AI-powered assistant can develop a deep understanding of your preferences and working style. This allows it to proactively suggest resources, tools, or actions that can streamline your workflow and boost your productivity.

portal can make reports, summaries and even a morning brief for you

• Conduct complex research in seconds, synthesizing information from across the web: With the ability to process vast amounts of data rapidly, an AI assistant can quickly gather and analyze information from multiple sources, saving you countless hours of manual research. It can identify relevant insights, summarize key points, and present the information in a concise, easily digestible format, enabling you to make informed decisions more efficiently.

portal can analyze data from csv files, use a python interpreter and even make plots

• Help you brainstorm ideas, solve problems, and achieve goals with unprecedented efficiency: By leveraging its natural language processing capabilities, an AI assistant can engage in dynamic, contextual conversations to help you generate new ideas, troubleshoot issues, and develop effective strategies. It can ask probing questions, offer alternative perspectives, and provide real-time feedback, acting as a virtual thought partner to enhance your problem-solving skills and creative output.

• Take care of routine digital tasks, freeing up your time and mental energy: An AI assistant can handle a wide range of mundane tasks, such as scheduling appointments, sending emails, organizing files, and filling out forms. By automating these time-consuming activities, it allows you to focus on higher-value work that requires human creativity, critical thinking, and strategic decision-making.

  Portal can book meetings - even multiple ones from the context of an email you're reading

The integration of advanced language models into our digital tools is not a distant dream but an inevitable reality. By bridging the gap between human communication and machine understanding, LLMs are poised to revolutionize the way we interact with technology, unlocking unprecedented levels of personalization.

Portal

A browser feels like the optimal interface for integrating AI systems into our everyday digital workflows. With the majority of people spending a significant portion of their computer time within a browser, it provides a familiar and accessible environment, with all the data an assistant needs readily authenticated and present.

With Portal, we've developed a shell for these models enabling them to navigate the internet just as you would – clicking links and buttons, scrolling through pages, and inputting data. The AI models powering this technology are roughly estimated to have the cognitive capabilities of a reasonably intelligent undergraduate student. And they are only getting smarter. Consider the boost in productivity from having a truly intelligent virtual intern dedicated to managing your digital interactions—previously a luxury only available to executives.

As these large language models improve, their expertise will increasingly approach that of specialized professionals, expanding their utility far beyond current capabilities. This is a really exciting time to be building in this space.

We're currently running a private beta, and we're excited to see how people use Portal to augment their workflows and explore the possibilities of this new kind of software. If you're interested in joining the beta, please sign up at beta.portal.so.

Privacy 

Portal stores everything in a local on-device database if you choose it, and we are figuring out the best way to do all of our inference locally. We’re betting that the kinds of open source large language models that can run on edge devices will keep getting better. Projects like ollama, llama.cpp, and llamafile, are pushing the boundaries for what’s possible in this space.

As AI becomes increasingly integrated into our daily lives and tools, the need for robust data privacy and ownership standards becomes more pressing. We believe that the development of a standardized, encrypted "context file" could provide a solution to this challenge.

The context file would serve as a secure, portable data container that encapsulates an individual's knowledge and preferences, allowing them to selectively share this information with trusted AI-powered services. By giving users full control over their data and enabling granular sharing permissions, we can foster a more privacy-centric and user-empowering AI ecosystem.

To make this vision a reality, we call upon the community of AI developers, privacy advocates, and standards organizations to collaborate on defining the technical specifications, encryption protocols, and best practices for our "personal contexts". If you are working or have any ideas in this space, please reach out.