Frank Addante's Hello Haven raises $15M for personal AI that remembers and acts

Why it matters

Hello Haven is betting the durable layer in consumer AI will be a user's memory and permissions, sitting above whichever model handles each task. That position could control the handoff between people, models and everyday apps.

Frank Addante, the repeat founder behind Rubicon Project, and engineer Duc Chau launched Hello Haven, a personal AI company, on September 17th with $15 million in pre-seed funding led by Mayfield.

Haven is their attempt to turn personal context into the operating layer for consumer AI. In a September 17th launch announcement, Hello Haven described the product as a "digital twin" that builds a private memory graph from a user's conversations, preferences, relationships, decisions and permissions. Haven uses that context to choose models and tools, then complete tasks across phone calls, texts, email, voice, the web and mobile apps.

The pitch grew from Addante's own struggle to keep up with the amount of information moving through his life. In a LinkedIn post about Haven's origins, Addante wrote that burnout and ADHD had left him struggling to decide what mattered. He first built Haven as a personal tool to organize his thoughts and reduce the noise before turning it into a product for other people.

That origin gives Hello Haven a more specific starting point than the usual promise of an AI assistant that schedules meetings faster. Addante is building for the executive-function burden behind the work: remembering why a task matters, preserving the decisions that shaped it and following through without requiring the user to reconstruct the context every time.

A memory layer with permission to act

Haven can prepare daily briefings, triage an inbox, send follow-ups, place orders, process returns and make calls, according to Hello Haven. Users can create recurring instructions in plain language, including requests for Haven to text or call before appointments. The same memory is meant to carry between channels, allowing a conversation that begins over text to continue by voice or email.

Hello Haven says its orchestration layer can draw from more than 500 AI models and technologies and connect with more than 100 applications. Those are company-supplied figures.

Hello Haven reported more than 2 million interactions since opening a private beta in August. The metric counts interactions, rather than individual users, paying customers or completed tasks, so it offers limited evidence about retention or commercial demand. Haven is free during the beta and requires no credit card, according to the product website.

Hello Haven announced the round without a valuation or naming additional participants.

Mayfield managing partner Navin Chaddha said, "Decades after backing Frank at Rubicon Project, we're excited to partner with him again at the inception stage of building this next wave of Personal AI." The reunion matters because Haven is asking users to place a large amount of personal history and decision-making authority inside a young product. Mayfield is underwriting founders who have already built and operated internet infrastructure at scale.

The ad-tech reunion

Addante's earlier companies included search engine Starting Point, email infrastructure provider StrongMail Systems and online advertising business L90. His Magnite biography says L90 went public before DoubleClick acquired it and that StrongMail was backed by Sequoia Capital. Addante later co-founded Rubicon Project, which built advertising infrastructure and became part of Magnite.

Chau brings the engineering half of that history. A 2020 Yieldmo announcement described him as a Rubicon Project co-founder and engineering leader who helped take the business public. Chau also architected the first version of Myspace as its lead software engineer and later served as CTO of Omaze and Yieldmo.

Their backgrounds explain the architecture Hello Haven is pursuing. Advertising systems route huge numbers of requests through data pipelines, identity systems and decision engines. Haven applies a similar orchestration problem to one person at a time, with models and applications replacing ad exchanges and publishers. The consumer promise is warmer. The underlying technical bet still depends on moving context between systems quickly and reliably.

Chau summarized that goal in the launch announcement: "The interface will keep changing. A person's context should not have to." Hello Haven plans to let developers build applications connected to a user's Haven. Hello Haven will also make a user's digital twin portable across televisions, cars, connected devices and robotics, according to the launch announcement. Those extensions remain a roadmap rather than capabilities demonstrated in the launch.

Trust is the product constraint

Persistent memory is already becoming a distinct product category. Personal AI markets editable memory stacks and personal language models, while Kin emphasizes encrypted, local-first storage. Inflection now presents Pi through Pi Journeys as an AI partner designed to grow with users across their lives. Memory infrastructure provider Mem0 sells persistent memory capabilities to developers building agents.

Hello Haven is pushing further into delegated action. A system that remembers a restaurant preference is useful. A system that places an order, sends a message or makes a call based on its interpretation of past behavior carries a different level of risk. Wrong memories, changed preferences and ambiguous permissions can turn personalization into an unauthorized action.

Hello Haven says each user's memory graph is private, owned and controlled by that individual. The strength of that promise will rest on practical controls: whether users can inspect and correct memories, restrict sensitive actions, understand which model received their data and reverse mistakes after Haven acts. For a product built around knowing its user, permission design will matter as much as model quality.

Addante's bet is that consumers will accept that responsibility in exchange for relief from the daily work of remembering, organizing and following up. His own experience with ADHD gives the product a clear reason to exist. The $15 million gives Addante and Chau room to prove that a digital twin can carry context across daily life without taking more authority than its user intended.