Claru

Claru is a training-data provider for robotics, embodied AI and world models, operated as the training-data platform of Reka AI Inc., the multimodal foundation-model lab whose legal entity is listed in Claru's own schema markup. Claru says its team built the data infrastructure behind Reka's vision models and Moonvalley's generative video models, and that it shares Reka's investors, naming General Catalyst, Khosla Ventures, Bessemer Venture Partners and NVIDIA. The company's site gives 2024 as its founding date; a third-party jobs profile says it launched in 2025 and that the core team works through Reka in San Francisco.

The business is built around real-world data capture rather than simulation or web scraping. Claru runs a network of more than 10,000 trained collectors in over 100 cities on six continents, each equipped with wearable cameras and following project-specific capture protocols, filming in settings such as home kitchens, factory floors, warehouses, workshops and city streets. Data types offered to robotics teams include egocentric first-person video that mirrors a robot head or wrist camera, manipulation trajectories with end-effector poses, gripper states and synchronised visual observations for imitation learning on arms such as Franka Emika and UR5 or custom humanoid manipulators, teleoperation demonstrations recorded with VR controllers, exoskeletons or leader-follower rigs on the customer's target embodiment, and navigation data with depth, IMU and odometry streams for mobile robots. Claru also collects game-environment, driving, cinematic and fashion footage for video-generation and world-model customers, and provides RLHF, video quality rating, pairwise model evaluation and red-teaming annotation.

Every clip passes through an enrichment pipeline that uses open foundation models including Depth Anything V2, ViTPose and SAM 3 to add depth maps, pose estimation, segmentation masks and structured metadata, after which expert human annotators label intent, context and edge cases. Data is delivered in the customer's format, and the company advertises going from brief to first delivery in days. Claru reports more than 4 million human annotations, over 100 active licensed datasets in its catalogue and partnerships with more than five frontier labs.

Published case studies include more than 500,000 egocentric videos captured for a world-modelling lab, 386,000 first-person clips collected through parallel GoPro, smartphone and activity-specific pipelines with same-day QA and weekly deliveries, workplace egocentric video recorded on site in ten business categories such as barista stations, carpentry shops and tailoring studios for a general-purpose robotics customer, more than 10,000 hours of synchronised gameplay data, and 976,000 human quality assessments used for RLHF by a video-generation lab. The site is available in English, Spanish and Portuguese, and the collector programme recruits contributors globally.