Replika’s founder built the AI assistant she pitched at Disrupt 12 years ago

Eugenia Kuyda launched Wabi for all users on Monday, opening her new personal AI assistant to anyone as a free download. The timing is pointed: twelve years earlier she had stood on the TechCrunch Disrupt stage presenting IO, a conversational lifestyle concierge built by her startup Luka that booked restaurants and vacations and handled other everyday tasks. IO predated the deep learning advances behind today's large language models, though Kuyda said it still worked and was pretty magical, while conceding it is nowhere near what is available now. Luka eventually became Replika, the AI companion company; Kuyda stepped down as its CEO last year and founded Wabi.
Wabi is built as a messenger app rather than a single chat window. Users keep several agentic threads, each devoted to a task or a long-term goal, and inside each thread they can build and use small custom apps. Examples from the company include a weight-loss tracker living inside a health thread, or an app listing nearby children's events inside a family thread.
The product has changed shape since its first release. When TechCrunch wrote about Wabi's initial launch a year ago, alongside a $20 million pre-seed round, it looked more like a social platform where users created mini apps from prompts and shared them with friends or across the app. Kuyda described the shift to Wabi 2.0 as an evolution rather than a pivot: the team concluded that people want personal software instead of one-size-fits-all App Store offerings, but often do not want to spend time building it themselves. In the new version the agent decides what to build. Connect it to your email and it spins up a set of starting threads and apps on its own; users can then ask for additional custom apps or open new threads.
Kuyda framed the combination as the differentiator against a crop of new personal assistants such as Muse and Instinct, which promise to manage inboxes, hunt flight deals and handle similar chores. Her argument is that Wabi does comparable assistant work while also carrying these apps, which helps users get more done. She also contrasted the multi-threaded design with single-conversation assistants, saying that one endless thread buries people in daily information. That concern has some support: researchers interviewed by TechCrunch note that the longer a chatbot conversation runs, the more likely it is to hallucinate, lose track of context or respond in ways safety guardrails would otherwise block. In practice, that is a familiar engineering problem, since a very long context window dilutes attention and makes retrieval of the right detail harder; splitting work across bounded threads is a plausible mitigation, though it moves the burden to whatever decides which thread holds what.
The multiplayer element is the second bet. Threads can be shared and behave like group chats, and participants decide how much to expose, from email and calendar data to bank account details. Kuyda's example: in a shared chat with her boyfriend, he can ask Wabi whether she is free next Thursday or whether she bought the tickets. Access is scoped to individual chats, and everyone in a chat can see when someone's personal information is requested. Her own shared threads include apps for live music, a daily relationship question, a running restaurant list, shared calendar events and a joint reading list with suggestions; a family thread tracks her daughter's schedule, finds kid-friendly events and quizzes her daughter on Japanese. Kuyda said today's agentic apps are mostly one-to-one conversations with an agent, and that she does not think anyone has cracked multi-player yet.
On security, Wabi is SOC 2 compliant, meaning its data controls have been independently audited, and it runs entirely in the cloud with most user data stored in the U.S., according to Kuyda. The permission model described is per-chat and visible to other members of that chat, which is a sensible default, but shared threads that mix several people's calendars, finances and family logistics create a genuinely hard consent problem: whoever joins later inherits visibility rules set earlier, and revocation has to propagate across every app built on top of the thread.
Wabi is free for now and the company has not settled on a business model. Kuyda said it is in talks to raise more funds. She linked the product back to her Replika work, which focused on AI that makes people feel less lonely, saying a degree of that carries into Wabi and that AI can bring people together, even if only to spend time better in real life, whether at a concert or over the next shared book.
Why it matters: People building assistants and consumer AI products now have a concrete rival that bundles agent-built micro-apps with shared, permission-scoped threads, so the comparison will fall on context handling and multiplayer consent rather than raw model quality. Anyone shipping agentic features will feel pressure to explain how long-running conversations stay coherent and how shared access is audited. Wabi's lack of a revenue model, and Kuyda's stated fundraising talks, suggest pricing and packaging are still open questions for teams watching this space, and my inference is that the free period is meant to buy usage data before any paywall appears.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Eugenia Kuyda's newest startup Wabi is positioned to take advantage of an idea she had more than a decade ago. Wabi offers a personal AI assistant with a twist, free and downloadable to anyone today.