Perpetual Automata is a machine learning research and product company working at the boundary between artificial intelligence and operating systems.

Artificial Intelligence has a permanent home on personal and work machines and will continue to act with more autonomy and access to private data. This shift toward non-determinism demands a fundamental advance in the observability and security inherent to every device. We believe the OS should detect, trace, and explain agent behavior beneath whatever framework you choose, as well as summarize actions, intent, and system interactions in plain language, with little to no engineering required on the part of the user. We aim to provide a solution with no API integration, no framework dependencies, and full flexibility to use whatever agent you prefer.

Our research spans agent detection and classification across local and networked environments, segmentation of agents from human users and conventional programs, language modeling techniques for translating low-level telemetry into human-readable explanations, extension of endpoint detection and response to the behavioral patterns unique to on-device agents, and methods for tracing agent intent from conversation context through tool invocation to system calls and resource utilization.