Useful help takes careful engineering.

Workflow

Understand a supported request, check the objects and workspace, ask for approval where needed, and stop when uncertain.

Required foundations

Appropriate hardware, privacy design, specific operating constraints, extensive validation, and a responsible human operator.

Scope limits

Unsupervised childcare, medical assistance, cooking, and other safety-critical care are outside our initial scope. No household capability is offered for purchase.

The hardest category, and the one most often oversold

Domestic and personal assistance is where public expectation runs furthest ahead of engineering reality. A home is unstructured in every dimension that matters: the objects are endlessly varied and often deformable, the layout changes daily, the lighting is uncontrolled, people and animals move unpredictably, and the tolerance for failure is very low precisely because the setting is intimate.

It is also the category where overclaiming does the most harm. A warehouse pilot that disappoints costs money. A household claim that disappoints erodes trust in the entire field, and in the worst case puts a vulnerable person in a situation a machine cannot handle.

Two capabilities that are routinely confused

Understanding a spoken request and executing it physically are separate problems with separate failure modes, and conflating them is the single most common error in this category. Language models have become good at interpreting intent. That competence says nothing about whether the machine can locate the object, grasp it without crushing it, carry it without dropping it, and put it somewhere sensible.

A system that understands fluently and acts unreliably is more dangerous than one that does neither, because the fluency invites trust the physical execution has not earned.

What a responsible narrow task looks like

The useful question is not which impressive scenario could be filmed. It is which narrow, supervised, genuinely helpful task could be done well, with a human responsible and present. Characteristics worth insisting on:

• A bounded request from a known set, not open-ended instruction following.
• Objects and a workspace the system has been validated against.
• Confirmation sought before acting where the consequence is not trivially reversible.
• Stopping when uncertain, as the default rather than the exception.
• A responsible human operator who is accountable and available.
• Privacy designed in from the start, because a machine in a home sees everything in it.

Explicitly outside scope

Unsupervised care of children, medical assistance of any kind, cooking, and other safety-critical care are outside this venture's initial scope and are not objectives of it. These are not areas where a venture should be learning in public. No household capability is offered for purchase, and nothing on this page should be read as a statement that a robot can safely perform any of them.

Privacy deserves its own answer

A device that perceives a home continuously raises questions that industrial deployments do not: what is recorded, what leaves the building, who can access it, how long it is kept, what happens to it when the device changes hands, and what a visitor who never consented is entitled to. Those answers have to be designed in, published, and defensible before anything is placed in anyone's home. They cannot be retrofitted as a policy page.

Current status

Concept only, and the furthest from reality of the three areas described on this site. Nothing has been built or tested, no hardware is supported, and no capability is available. Appropriate hardware, privacy design, specific operating constraints, extensive validation and a responsible human operator would all be preconditions, and none of them exists today.