Kali, the Data Company

← All posts

Datasets for Every Model

The Kali API beta is open. Here is what is inside it, how a request becomes a dataset, and what early partners are building.

For the last year, teams have come to us with the same problem in different words: they can describe the data their models need, but they cannot find it, collect it, or trust it at scale. Today that request path is open. The Kali API beta lets a team upload a specification and receive structured, reviewed data from the physical world.

What is available

The beta covers six collection programs: retail shelf conditions, warehouse and logistics workflows, light industrial assembly, laboratory bench work, food preparation, and outdoor maintenance. Each program returns video, time-aligned sensor streams, task boundaries, and annotations in a consistent schema.

Requests do not have to fit an existing program. If the environment is new, our field team scopes a collection protocol with you before anything is recorded.

From spec to dataset

  • Describe the environment, the tasks, and the labels you need.
  • We match the request to a collection program or design a new protocol.
  • Field teams capture the work with calibrated, synchronized equipment.
  • Every batch passes automated checks and independent human review.
  • Structured data arrives through the API, with provenance attached.

Timelines depend on how rare the environment is and how much review the labels require. Every accepted request ships with a schedule, a sampling plan, and a named field lead.

What partners are building

Robotics teams are training manipulation policies on warehouse and assembly footage. Operations groups are benchmarking vision systems against conditions in their own stores. Research labs are studying how people actually move through work, rather than how a script says they should.

The value is not more footage. It is footage that arrives with the context needed to learn from it.

Getting access

The beta is limited to a small group of teams so that every request gets a real collection plan. If you have a specification ready, or an environment you cannot find data for, we would like to see it.

Work with our data.

Join the waitlist →