MUUD: A Trust-Centered AI Operating System for Mental Wellness

MUUD Health is building consent-governed infrastructure underneath a mental wellness platform that combines a biometric wearable (MUUD Ring), a conversational AI companion, and human care networks, including peer support communities, licensed clinicians, and crisis resources.

Mental health technology today is a collection of parts, not a system. A wearable can measure your heart rate, but without context it can't determine what those signals mean for an individual's mental wellness. A chatbot seems like a solution, but it decides on its own what it remembers, what it acts on, and where your words end up, with no one accountable for that decision. A therapist sees you for fifty minutes and starts from nothing, because none of what came before reached them. Each piece works in isolation, and isolation is exactly where accountability goes to die: no one component is responsible for the whole person, and no person controls how their own data moves between the pieces that are supposed to be helping them.

We imagine a different dynamic. MUUD's answer starts with a single principle: no AI touches a member's data by default. Everything sensitive lives in the Vault, encrypted field by field with AES-256-GCM, and every access, not just every login but every read, every inference, every action, requires a durable, auditable consent record tied to that specific member. There is no standing access. An AI component doesn't get to hold a member's history because it might need it later; it has to ask, per action, against a record that can be checked and revoked. In practice, that means a member can see exactly which AI asked for what, when, and why, and can revoke that access as easily as they granted it. Consent isn't a checkbox at signup. It's the gate every access has to pass through, every time.

That same principle shapes how MUUD's AI is built, not as one general-purpose assistant with a master key, but as three narrowly scoped agents, each bound to a specific slice of data and nothing more. Each is intentionally designed with a limited scope, so no single system ever has unrestricted access to a member's complete history.

•      Infinity watches biometric and behavioral trends and surfaces what matters, without needing to see a member's journal entries or care network.

•      Studio helps a member make sense of their own journaled moments, and never leaves that lane to look at biometric or social data.

•      Harmony recommends people and community as trust builds, working only from the connection graph, not from a member's private history or journal content.

None of the three has broad access by design. If one is compromised, misbehaves, or is simply wrong, the damage is contained to what it was ever allowed to touch, and every touch is still checked against that member's consent record. Splitting capability this way isn't just an engineering choice; it's the same governance principle behind the Vault, applied to how intelligence itself is scoped.

This grant period is about making that architecture real everywhere it currently isn't. The Vault's consent-record system and per-field encryption are live and enforcing access today. The governor framework, Infinity, Studio, and Harmony, is not uniformly at that stage: Infinity's monitoring layer is further along, while Studio and Harmony are still being built and sequenced toward the same standard. The work ahead is to bring every surface, the Ring's firmware, the SAFE authentication layer, and the two governors still in progress, up to the same enforced standard the Vault already meets, and then to pilot the whole system with real members to find out where the boundaries hold and where they don't.