Twelve years ago this week, Eugenia Kuyda stood on stage at TechCrunch Disrupt and demoed IO, a lifestyle concierge chatbot that could book restaurants and vacations — before deep learning made such products workable. On Monday she officially launched the product that pitch was reaching for: Wabi 2.0, a multiplayer personal AI assistant now open to all users. Kuyda left her role as CEO of Replika, the AI companion company her startup Luka became, to build Wabi last year.
Wabi works like a messenger app rather than a single conversation. Each thread is dedicated to one task or long-term goal — health tracking, family scheduling, work projects — and inside each thread users can build and use custom mini-apps: a weight-loss tracker in a health thread, an app listing nearby children's events in a family thread, a shared restaurant shortlist between partners. Threads can be shared, functioning as group chats where several people interact with the same agent and see what data has been brought in.
The launch is a pivot from what Wabi was. The first version, launched last year alongside a 20 million dollar pre-seed round, was a social platform where users hand-built mini-apps from prompts and shared them with friends. Over the past year Kuyda says the team learned that while people love personal software over one-size-fits-all App Store solutions, they do not want to spend time building it. Wabi 2.0 flips the work to the agent: connect your email and it spins up a set of starter threads and apps on its own, then users add more by request.
The design argument is a bet against the single-thread assistant that Meta Muse and Instinct represent. Long conversations accumulate context until models hallucinate, lose track, or slide past their safety guardrails — a phenomenon documented in interviews with AI researchers. Splitting work into separate threads, Kuyda argues, keeps each one precise and relevant. The caveat deserves stating plainly: that is a company position, and no independent side-by-side test yet shows threaded assistants outperforming single-thread ones over months of use.
On privacy, Wabi says data access is scoped to individual chats — users choose how much to share per thread, from email and calendar details to banking information — and that the service is SOC 2 compliant with most user data stored in the United States. The company has not disclosed user numbers, underlying models or a business model; the product is free for now, and Kuyda says fundraising talks are under way.
What Wabi is really proposing is a new shape for the assistant wars. Every major consumer AI company is converging on one persistent assistant as the primary software interface, and the crowded field — Muse, Instinct, and the assistant features riding on ChatGPT and Gemini — mostly ships as one endless conversation with a shopping list of integrations. Wabi's answer is structural rather than cosmetic: many threads, many small apps, several people in the room. Whether multiplayer turns out to be the missing primitive or a niche preference will depend on whether shared threads get daily use — but Kuyda's bet is at least a clean, testable one: nobody, she says, has cracked multiplayer yet.
Comments (0)
Log in to join the discussion
Log InNo comments yet