Config Intelligence

Config Intelligence Inc., which operates as Config, develops data infrastructure and technology for general-purpose bimanual (two-handed) robotic manipulation. Rather than building robots, the company collects real-world human demonstration and motion data and converts it into training data structured for how bimanual robots move, selling that data alongside software and model products.

Its platform, Loop, comprises six composable modules spanning collection through evaluation, backed by two GPU engines, and turns a customer's real-world demonstrations into deployable, task-specific robot policies, with physically validated results fed back into training. CFG-1 is the company's first-generation robot foundation model for bimanual manipulation, trained on in-house human and robot data and described as adaptable to a customer task within hours. The company frames its approach as "closing the loop" between data and deployment, arguing that data is only validated by training models and testing them in the physical world, and works directly with customers on real-world deployment.

Config has also published research and datasets, including HABIT, a dataset for interactive action models aimed at robots that act alongside people and follow human direction, and work on vision-language-action models with long-horizon memory applied to tasks such as piano playing, shell games and object rearrangement.

Business model

A B2B model selling robotics training data together with the Loop platform and the CFG-1 foundation model to enterprises deploying robots; one source describes subscription-based services for ongoing support and access to proprietary technology alongside direct sales, with contracts covering deployment, maintenance and training, and plans to develop Robot-as-a-Service offerings.

Revenue is described as coming from partnerships and direct enterprise sales, with subscription-based support and technology access complementing direct sales of core systems.

Traction

Reported accumulation of more than 100,000 hours of collected human demonstration and motion data across roughly 300 staff in its collection studios, and over 10,000 training hours behind CFG-1. Product releases through 2026 include the HABIT dataset (August 2026) and research on long-horizon memory in VLA models (April 2026).

Latest developments

In April/May 2026 the company filed an SEC Form D disclosing a $33.3 million equity offering against a $35 million target, reported in press profiles as a $27 million seed round led by Samsung Venture Investment. In April 2026 it published work on robots coexisting with humans and on long-horizon memory for vision-language-action models, and in August 2026 it released the HABIT dataset for interactive action models.