Tim Berners-Lee invented the World Wide Web and chose to make it free and open. Decades later, the 70-year-old pioneer wants to shape the next layer of the internet. His latest effort? An AI agent called Charlie.
Built on Solid Foundations
Charlie isn’t some flashy new chatbot chasing venture capital. It emerged from years of quiet work at Inrupt, the company Berners-Lee co-founded to advance his Solid project for decentralized personal data storage. The concept dates back to 2017. Back then Berners-Lee described his desire for “an AI that worked for me. An AI which I could trust, and which I could share all my personal data, which would therefore be much more effective than an AI which I did not trust, and which I could trust more than an AI which would try to sell me things.” (W3C Design Issues).
Engineers at Inrupt turned that vision into a working prototype in late 2024. They built it around a simulated user named Bob. His Solid Pod held data pulled from sources like Strava running logs, bank records and fitness metrics. When Bob asked for running shoe recommendations, a generic LLM response demanded details on his pace, terrain and foot type. It offered safe but impersonal suggestions.
Charlie behaved differently. After Bob granted explicit consent, the system fed structured personal data directly into the model as context. The reply referenced his 5:29 pace, Green Lake Loop runs, 225 miles logged that year, morning routine and even preferred retailers. It suggested specific models from Nike, Saucony and Brooks. It noted his Chase Sapphire card for points and local Seattle fitting shops. “This is what I want,” Berners-Lee wrote of the contrast. The personalized version proved “devastatingly much more useful.”
But usefulness alone doesn’t define Charlie. Control does. The agent acts as gatekeeper. Your data stays in your Solid Pod. Charlie asks permission before sharing anything. When it does share, it obfuscates details to prevent the underlying model from building a complete profile of you. No raw data dumps. No permanent memory in some corporate vault.
John Bruce, CEO of Inrupt, spelled out the risks of the alternative. “People are already liberally uploading financial information to LLMs,” he told CNET. “The intimacy with which they then know you is frightening, and they never forget.” (CNET, June 3, 2026). Banks will likely introduce Charlie first. An app on your phone could follow.
This approach stands in sharp relief against today’s dominant AI assistants. Siri serves Apple. Alexa serves Amazon. ChatGPT, Claude and Gemini serve their creators’ commercial interests first. They ingest whatever users feed them. They remember. They optimize for engagement and revenue. Trust erodes when profit motives collide with personal privacy.
Berners-Lee addressed an audience at SXSW London this week. He remains “right chuffed” to see Charlie move from concept to reality after years of discussion about its values and operations. The inventor still believes making the web open was the right call. “It’s much more fun to have the web,” he said.
His timing feels deliberate. Major AI companies race toward IPOs. Valuations soar on promises of agentic systems that act autonomously on users’ behalf. Yet those same systems raise fresh questions about data ownership and consent. Recent coverage highlights the tension. One report explores how personal data wallets might prevent AI from absorbing everything users generate. Another positions Charlie as an evolution from current chatbots toward a true user interface for the web. (Raconteur, May 30, 2025; The New Stack, Sep 11, 2025).
Technical hurdles remain. Populating a real user’s Solid Pod with clean, up-to-date data from dozens of APIs demands significant effort. The 2024 prototype relied on simulated information. AI itself could accelerate that process by writing import tools. Data graphs built on semantic web standards would then mediate interactions between agents. The infrastructure matters as much as the intelligence.
Critics might dismiss the project as idealistic. Big Tech moves faster. Network effects favor platforms that already hold user data. Yet Berners-Lee’s track record commands attention. He created the web without patenting it. He has spent years pushing Solid as a counterweight to centralized data silos. Charlie represents the same impulse applied to AI.
Early demonstrations focus on practical scenarios. Health records. Financial advice. Personalized recommendations grounded in actual behavior rather than inferred profiles. Each interaction reinforces user ownership. The agent works for Bob. Not for an advertising network. Not for a model trainer. For Bob.
Industry insiders watch closely. Agentic AI promises to handle complex tasks across services. The winners will be those that solve the trust problem. Charlie offers one path. Store data under user control. Require explicit permission. Obfuscate when necessary. Use open standards for interoperability.
Success won’t come overnight. Partnerships with trusted institutions, starting with banks, provide the initial distribution channel. Momentum builds from there. Government bodies or health providers could adopt similar models. The vision extends beyond one assistant. It points toward an internet where individuals retain sovereignty over their information even as AI agents proliferate.
Berners-Lee never chased billionaire status. He chose fun over fortune when he gave the web away. That choice shaped the digital world we inhabit. Charlie tests whether the same principle can guide the AI era. Users decide if they want assistants that serve them. Or ones that serve someone else.
The prototype already shows the difference. Generic answers feel hollow once you’ve seen tailored ones drawn from your own life. The gap isn’t small. It’s decisive. And it arrives at a moment when many grow weary of surrendering data for marginal convenience.
So the question lingers. Will enough people demand control? Will institutions embrace systems that limit their own visibility into user behavior? The answers will determine whether Charlie remains an interesting experiment or becomes the template for personal AI.
Berners-Lee sounds optimistic. After all, he built the original platform. Now he’s helping define what runs on top of it.
WebProNews is an iEntry Publication