Node Data
Kids film their chores. Robots learn to do them.
Robots are moving into people’s houses. But almost every video used to train them was filmed by adults — no kids around, no mess. A robot that has only ever seen a research bench or a perfectly tidy house doesn’t see the world the way kids do.
A household robot is going to spend its whole life around children. It has to handle how we move, what we leave on the floor, and what a room looks like from three feet up. Adult footage from a lab has none of that.
So kids are going to fix it.
Node Data is the first physical AI data company run by kids, for kids. Here’s how it works: a kid opens the ND GO app, picks a chore — tidy their room, load the dishwasher, put the laundry away — films themselves doing it, and earns $1.00 — real money, paid into their parent’s account. Parents stay in charge the whole time: they say yes before the camera ever turns on, and they can see every clip.
Almost nobody else can collect this data. Filming children needs a parent’s consent, and a robotics company with no relationship to families has nobody to ask. We built the consent architecture first, so the families we work with already know exactly what they are agreeing to.
The vision
Eventually, every household will have a robot in it.
Node Data’s job is to make sure robots understand kids and can interact with them.
The questions we are built on
- 01
Why kids, and not adults with better cameras?
Because children are who a household robot will be working around, and because the house looks different from their height. A model that has only seen adult hands at adult eye level meets both of those for the first time in a customer's kitchen.
The app they film with ↗ - 02
Why can't anyone else collect this?
Filming a child takes a named parent's permission, and a robotics company with no relationship to families has no one to ask — which is why this footage is missing from every open dataset. We built the consent path before the camera, so a parent can see and withdraw every permission separately.
The eight rules we hold ourselves to ↗ - 03
You already have training data. What is this for?
Refinement, not bulk. Your base model is trained; it fails in particular ways once it is in a real house. You name the failure, it becomes a bounty, and households film that exact task — so what you buy is the part your dataset is missing rather than more of what it already has.
Describe what it gets wrong → - 04
Who decides a clip is good enough?
Code does, not a mood. Every clip is scored against 4 criteria published before you spend anything, at the moment it lands. A clip that fails is never delivered and never charged.
The acceptance criteria → - 05
Who said yes before the camera turned on?
A named adult, on the record. Recording, sharing with a buyer, and training AI are three separate parental permissions, each withdrawable, each checked again on every single clip rather than once at signup.
What a parent controls ↗ - 06
What is one clip actually worth?
$2.50 from the buyer, held in escrow until the clip is accepted, and $1.00 of it to the household that filmed it. Whatever nobody captures is refunded.
Post a bounty →
The rest of Node Data
- Custom dataPay for clips of a task you need
- MarketplaceDatasets and policies for sale
- Docs & SDKTyped client, every endpoint
- AI FactoryDraft a dataset from a prompt
- Developer programFree, no account
- EnterpriseVolume, licensing, support
- PricingFree, creator, enterprise
- NewsWhat shipped, and why
- For familiesSet up a household on nodedata.ai
psst — press ⌘K. or the up arrow. twice.
