A vision originally pitched more than a decade ago has finally materialized into a full fledged consumer platform. Eugenia Kuyda, who previously co-founded Luka and led Replika, has officially opened her new startup Wabi to the public. Designed as a multiplayer personal AI assistant, Wabi operates as an interactive messenger where individuals can create and maintain multiple dedicated threads for separate life projects, household logistics, and long-term personal goals. Inside every conversation, autonomous agents build and run lightweight customized applications on demand.
The Evolution from Luka and Replika to a New Agentic Vision
Kuyda began exploring conversational software around twelve years ago with IO, the debut product of her startup Luka. Launched in New York City as a natural language concierge for discovering and booking dining venues, IO operated well before the contemporary revolution in deep learning architectures and massive generative language models. While that early system demonstrated notable functional charm, the underlying software capabilities of the era fell far short of modern intelligence systems.
Luka subsequently underwent a major strategic pivot, transforming into Replika, a widely recognized virtual companion platform created to alleviate loneliness through empathetic digital interactions. Last year, Kuyda stepped down from her position as chief executive officer of Replika to start Wabi, backed by a twenty million dollar pre-seed financing round. The startup originally envisioned a community ecosystem where users would construct lightweight applications via natural language prompts and distribute them publicly across a shared network.
Shifting to Wabi 2.0: Letting the Agent Handle Software Creation
User interactions over the past twelve months revealed an important operational insight: while consumers clearly appreciate personalized software over generic, rigid offerings found in mobile app repositories, they have little desire to invest time and effort into building those tools themselves. In response, Wabi shifted its approach into what Kuyda describes as an evolution rather than a complete departure.
Under the freshly unveiled Wabi 2.0 framework, the agent takes full ownership of the software creation process. Once an individual grants inbox access, the system autonomously analyzes daily workflows to generate relevant conversational threads and initiate useful mini apps. Users can subsequently expand their setup by requesting specialized micro software or launching fresh conversational topics. For instance, a dedicated health channel can feature an automated body weight monitor, while a domestic household channel can integrate a radar for regional youth activities alongside educational practice drills.
Preventing Context Collapse Through a Multithreaded Architecture
A primary architectural distinction separating Wabi from contemporary single stream AI helpers lies in its multithreaded layout. Conventional assistants typically channel every request, reminder, and query into one never-ending conversational thread. Over time, that singular record accumulates immense context volume, leaving users overwhelmed under an avalanche of daily notifications and unstructured logs.
Prolonged single thread conversations also introduce serious technical drawbacks. AI researchers point out that as conversational history stretches indefinitely, models become increasingly prone to memory degradation, hallucinated responses, and behavioral drift that can bypass intended safety safeguards. By isolating distinct workflows into independent channels, Wabi keeps the contextual working memory focused, preventing unrelated chores from confusing the underlying agent.
Collaborative Multiplayer Channels for Friends and Families
Beyond compartmentalized threads, Wabi introduces a multiplayer dynamic aimed at breaking down the isolated one-on-one nature of current AI assistants. Rather than restricting artificial intelligence to isolated individual siloes, the platform enables families, couples, and project teams to interact with autonomous agents collectively within shared group chats.
Shared communication spaces allow members to embed collaborative widgets along the top of the interface. In personal settings with a partner, these tools can encompass real-time concert trackers, shared calendar syncs, curated restaurant registries, daily relationship prompts, and reading lists that offer mutual book recommendations. Within family environments, threads can coordinate extracurricular timetables, scout local weekend gatherings, or provide language learning exercises. Group participants can openly query the agent regarding practical logistics, such as checking a partner's calendar availability or verifying ticket purchases.
Data Privacy Controls and Future Expansion Plans
Entrusting personal communication channels, emails, and financial data to autonomous agents naturally raises privacy and security concerns. To address this, Wabi implements strict contextual boundaries where data visibility is scoped specifically to individual channels. Every member inside a shared thread receives explicit visibility whenever private information is accessed or queried by the AI assistant.
On an infrastructure level, the company has attained SOC 2 compliance following independent third-party audits of its internal data handling safeguards. Wabi operates completely on cloud infrastructure, with the vast majority of consumer information hosted within United States servers. Whereas Replika focused heavily on offering synthetic emotional companionship, Kuyda highlights that Wabi aims to use digital intelligence to facilitate richer interpersonal interactions in the physical world. The service is currently offered free of charge while leadership evaluates potential monetization avenues and prepares to secure additional venture investment.
