About Praxis · Y Combinator S26
Building the collection layer for embodied AI.
Praxis is a robotics company. We do not build robots — we build the thing every robot company is short of. Our network turns the task footage generated by existing workforces into annotated training data for frontier AI labs, and pays the businesses that host collection a share of every contract we sign.
Four goals, in order of how far out they sit.
The near-term business is a revenue share on annotated task data. The long-term business is the network that produces it.
Make real-world task data abundant
Every frontier lab building humanoids is rate-limited by the same input: diverse, high-fidelity footage of people doing real work. That footage exists — it is being generated on millions of shifts a day and thrown away. Praxis exists to capture it, structure it, and put it in front of the labs that need it, at a scale no purpose-built collection studio can match.
Pay the workforce that already exists
The cheapest way to collect task data is not to hire people to perform tasks — it is to instrument people already performing them. That is why our economics work as a revenue share: the partner business carries no cost and no operational change, and we can outbid dedicated collection on cost per verified hour while paying more for the data.
Become the standing collection network
Each partner adds geography, industry, and task variety — and diversity is the largest multiplier on dataset value. A network spanning countries and sectors is worth more than the sum of its sites, and it is not something a competitor can assemble quickly. That network is the company.
Give the data back as internal tooling
The same annotated footage that trains robots describes how a business actually runs. Digital twins, surfaced safety risk, and training curricula built from the work of your best operators are all downstream of one capture pipeline. Partners get the royalty and the tooling.
The commitments that make a partnership signable.
Never trade secrets
We do not film IP or proprietary process, and we delete footage on request. A partnership that requires a legal fight over scope is not a partnership we want.
Privacy by default
Faces, screens, license plates, and documents are automatically detected and redacted before storage — not flagged for later review.
Aligned economics
We only earn when accepted data is delivered, and partners are paid on verified hours. Nobody is paid for footage the labs reject.
Industry and country agnostic
All data is valuable. Task footage trains general-purpose robots, not task-specific ones, so breadth is a feature rather than a dilution.
Who you will actually be working with.
Four operators run the company. Open any card for background.