- Specialization
- Robotics
- Business
- Data Provision
- Market
- Manufacturing
- Ecosystem
- Physical AI
- Customers
Enterprisesand 1 more customers valuesLess
- Researchers
More details
- Organization type
- Commercial Company
- Ecosystem
- Physical AI
- Business
- Data Provision
- Specialization
- Robotics
- Market
- Manufacturing
- Customers
- Enterprises, Researchers
- Delivery models
- Professional Services
- Revenue models
- Product Sales, Services Fees
- Listing status
- Not classified
- Latest equity round
- Seed
- Reported funding amount
- Not classified
About
Cortex AI builds large-scale real-world workplace robot and egocentric datasets, including human video with pose, depth, subtask labels, and robot trajectories. It sells data and related collection and evaluation access to frontier labs developing robotics foundation models and general-purpose robots; its differentiator is industry-scale, in-the-wild data gathered through real workplaces and human-in-the-loop workflows.
Market
Cortex AI competes in embodied-AI and physical-robotics data infrastructure, supplying the real-world datasets and evaluation operations needed to train and validate robotics foundation models. Its differentiation is an integrated pipeline spanning workplace egocentric data, embodiment-specific robot trajectories, human-supervised deployments, recovery data, and a marketplace of real workplaces; competitors such as Sensei, DataX Power, Trossen Robotics, and Claru are more focused on particular combinations of demonstration capture, teleoperation, annotation, or managed data collection.
Ideal customers are frontier robotics labs and robotics companies—especially teams developing humanoid, general-purpose robots, and robotics foundation models. Primary buyers are robotics research and engineering leaders who need real-world pretraining or fine-tuning data, deployment evaluations, and recovery trajectories.
At a Glance
Problem
Embodied-AI and robotics companies face a data bottleneck: training general-purpose robots requires large, diverse datasets from real workplaces rather than only lab or synthetic environments. The pain is operational and potentially expensive because models must learn how people perform tasks, how robots behave in industrial settings, and how to recover from failures before they can be deployed reliably. Cortex AI’s central use case is collecting factory- and workshop-based video and robot trajectories that robotics companies can use to train and fine-tune their models.
Product / Service
Cortex AI operates a real-world data-collection and deployment network. Through industry partners, it captures egocentric human video with hand and body pose, depth, and subtask labels, as well as trajectories from manipulators and humanoid robots. It also supports human-in-the-loop rollouts in which remote operators recover robots when they fail; those interventions become additional training and evaluation data. The company sells this data to robotics companies, providing embodiment-specific material for fine-tuning foundation models and improving performance through repeated real-world feedback.
Market
Cortex AI competes in the emerging embodied-AI data and robotics-infrastructure market, serving frontier labs building robotics foundation models and general-purpose robots. Its category overlaps with providers of teleoperation data, reinforcement-learning environments, robot trajectories, and real-world evaluation services. The available research does not identify named direct competitors, although the broader robotics market includes companies offering robot operation, teleoperation-data collection, and reinforcement-learning environments.
The company appears to have early commercial and technical validation: its materials describe a network of industry partners, an offering that sells data to robotics companies, and involvement in enabling the MolmoAct2 robot foundation model. However, the available evidence does not disclose revenue, customer counts, funding, or deployment volume, so Cortex AI’s financial traction and whether it is profitable or pre-revenue cannot be determined.
Founders & Leadership
Funding History
500 Global, Golden Gate Ventures, Matrix Partners, Liquid 2 Ventures, Y Combinator, Gokul Rajaram, Kaz Nejatian
Recent News
Cortex AI announced Cortex Harness, a platform for running AI models on real robots with remote human guidance. Operators can steer robots with prompts or take control, while the platform collects interventions as recovery data; the company invited researchers, developers, and industry partners to join its beta program.
The Allen Institute for AI announced MolmoAct 2, an open robotics foundation model designed to improve 3D action reasoning for real-world robot tasks. Cortex AI’s official site credits the company with enabling MolmoAct2, making this the clearest documented integration involving Cortex during the period.
In this interview, Lucas Ngoo explained Cortex AI’s focus on collecting large-scale human manipulation and robot-training data rather than building robots or training the models themselves. He described Cortex as the data layer for general-purpose robotics, with teleoperation and real-world deployment planned as the company scales.
DroidAge profiled Cortex AI as a San Francisco robotics company providing real-world workplace robot datasets and egocentric video with hand/body pose, depth, and subtask labels for training robot AI systems.
Cortex AI launched a platform for real-world egocentric and robot data, including workplace data collection, robot trajectories, human-in-the-loop rollouts, and the Cortex Marketplace for hosting capture and teleoperation sessions. The company says its data is intended to support robotics foundation models and general-purpose robots.
Active Roles
4Business Model
Cortex AI monetizes by selling packaged real-world robot and egocentric data, along with related data-collection and evaluation services, to robotics companies and frontier AI labs; reported paid contracts support this B2B model. Its Cortex Marketplace also enables workplaces to host paid data-collection and evaluation sessions while labs access the resulting in-the-wild data.