Agent security without the security team
If we've done our homework, you fall into one or both of these buckets:
Either way, you need a quick solution with minimal upkeep. No security background needed, no security team needed.
We built Perpetual Automata to be that solution. One line to install. Immediate visibility into agentic behavior on all your devices. We catch and stop misaligned agents before they can do damage.
You're a developer using Claude Code, Cursor, or Codex to ship faster. These agents have elevated access to your codebase, your credentials, your filesystem. They can be compromised with a single sentence — prompt injection, memory poisoning, tool poisoning. You have no way to know if an agent is doing exactly what you asked or something else entirely.
Your org has a different problem. 78% of AI users bring their own tools to work. Developers are installing agents on their laptops right now — with access to databases, APIs, and internal systems. IT doesn't know. Security doesn't know. There's no oversight, no audit trail, no line of defense once an agent is compromised. If your org hasn't banned these tools ouright, thye've either locked them in restrictive sandboxes or turned a blind eye and hoped for the best. Neither is a real solution.
Everyone wants to embrace agentic tools that 10x their productivity, and we want to enable this without sacrificing device security.
Every action an agent takes — file access, network calls, code execution — must pass through the OS kernel. This is the one layer an agent cannot escape or manipulate. We sit beneath the agent, monitoring and restricting behavior regardless of whether the agent is compromised.
Think of it as the airplane black box for on-device AI: tamper-proof, always recording, completely independent of the system it monitors.
One line to install. Immediate visibility into what every agent on your machine is actually doing. No configuration required to get basic security. Natural language to configure advanced policies.
Colton Botta (CEO/Co-founder): BS Computer Science and Entrepreneurship (NC State), Valedictorian. 2021 Marshall Scholar. MSc Cognitive Science (Edinburgh), MPhil Machine Learning Research (Cambridge), PhD Generative AI (Edinburgh). Former software engineer at Pendo.
Jake Henderson (CTO/Co-founder): BS Economics and Statistics (NC State). Deep expertise in distributed systems, security, and data infrastructure. Built distributed systems at IBM Cloud (stream processing at global scale), SAS Institute (secured data ontology graphs for governments, banks, and law enforcement), and Rail Inc.
We're hiring exceptional engineers who want to work on hard problems in AI security, and be at the forefront of enabling agents to run safely on all device types.
Interested? Or want to pitch us another role you think we need? Contact us below