ADELAIDE, Australia -- Australian artificial intelligence startup Metacognition has raised $10 million in an early-stage funding round ed by Main Sequence to develop what it describes as an operating system for robots and AI agents.

The software is intended to sit beneath large language models and control how they interact with machinery and other systems. Metacognition says its approach could allow people without robotics expertise to train hardware through spoken instructions and corrective feedback.

The company has not disclosed a commercially available product, customer deployments or independently evaluated performance results. Its reported development work includes simulations in which operators can direct an open-cut mine or manage an electricity grid using voice commands, according to Startup Daily and Dealroom.

Metacognition was founded in 2026 by three scientists from the Australian Institute for Machine Learning and Australian National University. Main Sequence, a venture capital firm backed by Australia’s national science agency, CSIRO, led the funding round. 

The startup’s website says its technology focuses on persistent memory, continual learning and reasoning across “always-on” AI systems. In robotics, those capabilities could help a machine retain information from earlier tasks and apply corrective instructions over time.

Metacognition plans to target existing industrial equipment before pursuing broader robotics applications. Mining is among the sectors identified by the founders, where operators already use autonomous and remotely controlled machinery but face strict safety and reliability requirements.

Metacognition has not announced product pricing, a release date, paying robotics customers or results from tests involving physical robots.