Solutions

Intelligence proposes. Sentinel authorizes.

Every U1ai product sits behind the same boundary: models can interpret and recommend, but actions require authority that no amount of model confidence can mint.

The products

Built on one boundary

Myelin Live

An app for the neurodivergent and ADHD community: a user-set Friction Dial, micro-steps, a Focus Mode timer, and the Body Double companion for talking, journaling, study, and breaking down complex terms. Runs on-device.

myelin-u1ai.vercel.app →

Sentinel & Glass Box Open source

The authority gate for agent actions - ALLOW, REVIEW, or DENY, where a deny cannot be lifted by model confidence - plus the evidence layer that keeps what was said, what was interpreted, and what is proposed distinct.

github.com/Unmute1-Ai/glass-box →

Primer In development

A child-first, accessibility-native AI learning environment. Intelligence recommends; authority decides. Success is measured by the child becoming more capable without the AI - not by engagement.

True2u Companion First pass

A neurodivergent companion app. “Make today smaller”: type a messy thought, split it into steps, edit before saving. Early prototype.

Under the glass

How Sentinel decides

Pre-execution governance: a proposed action is checked against signed authorization envelopes, explicit policy, and state-based execution control before anything runs. Authorized actions pass; denied or ambiguous ones escalate to a person.

Sentinel runtime authority and execution control architecture: proposed actions from AI agents pass through the Sentinel pre-execution authorization layer, which returns an authorized or denied/escalate decision before reaching execution targets.
Sentinel runtime authority & execution control - from the U1ai design docs.