Eugenia Kuyda launches Wabi 2.0, a multiplayer AI assistant built on threads

Twelve years to the day after Eugenia Kuyda pitched a lifestyle concierge chatbot called IO on stage at TechCrunch Disrupt, she has launched the product her original startup promised: Wabi 2.0, a multiplayer personal AI assistant that opened to all users on Monday.

Kuyda's first company, Luka, built IO before the deep learning breakthroughs that produced today's large language models (AI systems trained on massive text datasets to generate conversational responses). The technology failed to keep up with the vision then. It has now caught up, and the market has become crowded with personal AI assistants such as Muse and Instinct that promise to manage inboxes and find flight deals.

"It still worked," Kuyda told TechCrunch in an exclusive interview about the 12-year-old IO demo. "It was pretty magical…but, of course, it's nowhere near what we have today."

Luka eventually became Replika, the AI companion chatbot company. Last year, Kuyda left the Replika CEO role and founded Wabi, which raised a $20 million pre-seed round reported by TechCrunch at the time. The stakes are straightforward: every major consumer AI company is now betting that a single persistent assistant becomes the user's primary software interface, and Wabi is arguing they are all building it the wrong way.

What changed between Wabi 1.0 and 2.0?

The original Wabi, launched last year, was closer to a social platform where people created mini apps with prompts and shared them with friends. Over the past year, Kuyda said, the team learned something about its users: they love personal software rather than "one size fits all solutions in the App Store," but they "maybe don't want to spend a lot of time, or any time, building it."

Wabi 2.0 flips the work onto the agent. The moment a user connects it to email, the agent automatically generates threads and apps to start with. Users can request custom apps or start new threads on top of that.

Each thread is dedicated to a task or long-term goal, and inside each one users can build or use custom apps — a weight-loss tracker inside a health thread, for example, or an app showing nearby children's events inside a family thread. Wabi functions as a messenger app, so these threads sit alongside each other rather than piling into a single conversation.

"It's doing what Instinct and Muse do, but it also has all these apps, which helps users do more stuff," Kuyda said.

The takeaway: Wabi now asks users for connections to their data, not for their time building software.

Why does Kuyda think single-thread assistants fail?

The design bet separating Wabi from its competitors is multi-threaded and multiplayer. Kuyda argued that rivals' single-conversation models degrade with use.

"It's not just everything in one single thread like with Instinct and Muse, where you get buried in the information that it sends you everyday," she said. "Single thread experiences, if you think about it, get worse over time because it's just so much context in one endless conversation."

That claim has some research support. According to interviews TechCrunch conducted with AI researchers, the longer a chatbot conversation runs, the more likely the model is to hallucinate (generate false information), lose track of context, or respond in ways its safety guardrails would otherwise block. Splitting an assistant's work across threads is a structural answer to that problem, not just a UI preference — though Wabi has not published benchmark data showing its thread model measurably reduces such failures.

Threads can also be shared, functioning like group chats between people and the agent.

"Today, all the agentic apps are just [conversations] between you and the agent, and I feel like it should be a lot more about bringing people together in a group chat together," Kuyda said. "Not you doing it separately with your AI, and I'm doing it separately with my AI…I don't think anyone has yet cracked multi-player."

The takeaway: Kuyda's argument is that context overload is a technical problem, and separate threads are the fix.

How does sharing work in practice?

Kuyda walked TechCrunch through her own shared threads, which serve as the product's clearest demonstration. In a thread with her boyfriend, a row of apps along the top includes:

• A live music tracker
• An app that asks a daily question about their relationship
• A running list of restaurant suggestions
• Important events from their shared calendar
• A list of books they are reading together, with suggestions for what to read next

Her family thread holds an app tracking her daughter's schedule, one for finding kid-friendly events nearby, and one that quizzes her daughter on Japanese.

Within any shared chat, users can share as much or as little personal data as they choose — from email and calendar information to bank account details.

"So my partner can ask Wabi in the chat, 'Is Eugenia free next Thursday?' or 'Did she buy those tickets?'" Kuyda said. She noted that access is scoped to individual chats, and everyone in a chat can see when someone's personal information is requested.

That visibility rule is a guardrail, not a guarantee. The company has not detailed what audit logs exist, how revoked threads are handled, or what happens to shared bank data if a shared chat dissolves. Those questions will matter more as Wabi asks users to connect financial accounts.

The takeaway: shared threads give a household one assistant instead of several parallel ones, with visible-but-narrow controls.

What about privacy and data storage?

Wabi is SOC 2 compliant, meaning an independent auditor has reviewed its data security controls. The app runs entirely in the cloud, and most user data is stored in the U.S., Kuyda said. She did not specify where the remainder is stored or which vendors process the data — a gap worth noting for an app ingesting email, calendars and bank information.

Kuyda framed the multiplayer direction as a continuation of her work at Replika, but with a different emotional goal.

"My previous company Replika worked on AI that will make people feel…less lonely," she said. "To a degree, that is [in Wabi] as well. AI can bring people together, even when it comes to just helping you spend time better together in real life, whether it's going out and seeing a concert together or finding the next book that you will read together."

The takeaway: Wabi meets a formal compliance bar, but the company has disclosed little about its downstream data handling.

What is the business model?

There isn't one yet. Wabi is free for now, and the company hasn't settled on how it will make money. Kuyda said Wabi is in talks to raise additional funds, on top of the $20 million pre-seed from last year.

That leaves two open questions the launch does not answer: whether consumers will hand an agent their email and bank details en masse, and how Wabi monetizes without eroding the trust that data-sharing requires. Kuyda has a 12-year track record of being early to consumer AI — Replika proved the demand for AI companionship — but being early to multiplayer agents means proving the model before the incumbents adopt it.