Wabi Raises $55M to Build Mini Apps Inside a Group Chat

Wabi, the personal AI agent built by Replika founder Eugenia Kuyda, has raised $55 million at a $1.25 billion valuation, the company announced on Thursday. IVP led the round, with Andreessen Horowitz, Forerunner, the Amazon Alexa Fund and a long list of angels joining.

The money is not going into a bigger chatbot. It is going into an agent that writes its own interfaces.

What Wabi actually ships

Wabi started life as a prompt-to-app builder. In September it pivoted to a messenger, and the current product combines an AI agent, personalized mini-apps, and shared spaces where several people work with the same agent.

The mini-apps are the interesting part. A couple keeps a running list of restaurants to try and checks them off together. A family turns a weekly meal plan into a grocery list. A book club gets a tracker, a group of friends gets a trivia game for a birthday. Each tool can be reopened and adapted later, so it behaves less like a one-off chat answer and more like a persistent artifact.

Why investors paid up

The round values a company that only recently opened its product publicly, and the cap table reads like a consumer-AI endorsement: Shopify's Tobi Lütke, Salesforce's Marc Benioff, Marissa Mayer, Dick Costolo and Alex Pall all put money in.

"AI shouldn't just save people time. It should help them spend it better, doing more of what they love, with the people they love," Kuyda said. IVP's Alex Lim framed the bet as moving past the interface itself, arguing that Kuyda has "a clear vision for how people will interact with AI, one that goes far beyond the text box or command line."

The comparison set explains the price. Instinct reportedly hit a $10 billion valuation in September and Tolan raised $30 million. Investors are paying for distribution and habit in consumer agents, not benchmark scores.

What it means for builders

For anyone shipping AI automation, Wabi points at a shift worth watching: the interface is becoming a generated object rather than a fixed product surface.

That changes what a builder's real deliverable is. If the agent can compose a small tool per task, the durable assets are the tool layer, the data model, the permission rules and the shared context - not the screens. Teams that treat prompts as the product will find the UI evaporating around them; teams that expose clean, composable actions to an agent inherit the UI for free.

Multiplayer agents also drag in a problem single-user assistants dodge: when three people share one agent and one generated tool, whose memory applies, who can edit the plan, and what gets logged? Wabi's shared spaces are a permissions model wearing a consumer-friendly face.

The open questions

Retention is the one to watch. Generated mini-apps are easy to make and easy to abandon, and a $1.25 billion valuation prices in the assumption that people come back the next week, not just the day the tool was made. Monetization is unstated, and the strategic risk is the usual one: when the assistant becomes the surface, whoever owns the underlying models sets the terms.