On Tuesday, principal software engineer Sally O’Malley from Red Hat unveiled a new open source utility named Tank OS designed to facilitate the safe deployment and management of OpenClaw agents.  

“This was a fun project that I assembled over the weekend, and I realized it would be a great fit for AI and our future direction,” she shared with TechCrunch, emphasizing her desire to distribute it “to the masses.”

Tank OS is tailored for advanced users wanting to operate OpenClaw on their own systems as well as IT professionals overseeing multiple corporate OpenClaw agents. It enhances the safety and simplifies the maintenance of OpenClaw in bulk.

Numerous individuals, companies, and startups are already innovating better methods to utilize OpenClaw — the open source initiative that deploys an AI agent onto a local machine. A rising number of startups are also developing competing alternatives to claw that they claim are more secure (for example, NanoClaw). 

What sets O’Malley’s initiative apart is her role as an OpenClaw maintainer. This indicates that she is among the few software engineers collaborating with creator Peter Steinberger to determine which features and bugs are prioritized. Her focus is on improving OpenClaw’s performance in enterprise applications, along with various versions of the Linux operating system from Red Hat. (Although Steinberger was recruited by OpenAI, he continues to lead the independent open source initiative OpenClaw.)

O’Malley became involved with OpenClaw because she perceives it as a way to “enable everyone to run AI in a secure, open manner,” she stated. 

However, she began contemplating the implications of OpenClaw entering an enterprise environment and resolved to create a tool for that scenario. She started with an open source container solution known as Podman, developed by a peer at Red Hat. Containers allow applications to run independently from the host computer, with all necessary components bundled together. For instance, they can enable a Linux application to operate on a Windows or Mac system. 

Podman is particularly secure since it operates in a “rootless” manner, meaning it does not grant the containers any privileges from the underlying machine, according to Red Hat.

Tank OS installs OpenClaw onto Red Hat’s Fedora Linux operating system within a Podman container and transforms that container into a bootable image, so it automatically executes and initiates OpenClaw upon system startup. 

Her tool encompasses all elements necessary to make OpenClaw functional without human supervision, such as state (which enables memory retention); the capability to store API keys (credentials for accessing subscriptions and services); along with other functionalities.

Users can execute several Tank OS instances on one machine for varied tasks, ensuring that passwords or credentials are not shared, and no OpenClaw instance can access other operations on the system. 

While O’Malley acknowledges that the OpenClaw initiative is striving to enhance agent safety, she remarks that “it’s an incredibly powerful application,” but it can also be “hazardous” if not correctly configured. “It’s not a tool that’s easy to use unless you possess some technical expertise,” she noted. 

Numerous accounts exist, such as the Meta AI security researcher whose Claw started erasing all her work emails, or an agent that downloaded all of a user’s WhatsApp direct messages in plain text. There is also an increasing amount of malware targeting OpenClaw users. 

It should be noted that Tank OS isn’t designed for beginners either, she notes. Users must be comfortable with installing and maintaining software on their devices, she stresses. Tank OS isn’t the only OpenClaw deployment within containers; for example, NanoClaw is undertaking a similar endeavor utilizing the well-known container technology Docker.

Nevertheless, Tank OS aims to be particularly advantageous for IT professionals (essentially, Red Hat’s primary clientele) who may eventually oversee a multitude of OpenClaw agents across corporate systems. It allows them to update agents in the same manner they already manage other containers. 

“My involvement with OpenClaw reflects my genuine interest in it,” O’Malley expressed. “I’m excited about how it’s going to evolve when millions of these autonomous agents communicate with one another.”