Sally Ann O’Malley is a principal software engineer in Red Hat’s Emerging Technologies organization and the creator of tank-os, an open-source prototype that runs AI agents on a hardened, image-managed Linux operating system. She applies enterprise Linux security, container orchestration, and observability to a practical challenge: making autonomous agents portable, recoverable, and safe to operate at scale.
O’Malley joined Red Hat in 2015, working on Linux security, containers, Kubernetes, and OpenShift before moving into Emerging Technologies. Her 2016 guide to developing against OpenShift in a virtual machine anticipated her continuing interest in reproducible local and remote environments. At DevConf.CZ 2024, she co-led sessions on container fundamentals and Podman-based image builds across different processor architectures.
In 2025, she coauthored a local RHEL command-line assistant for disconnected environments and developed guidance on observability for distributed LLM inference. The latter emphasized measurements conventional infrastructure dashboards often miss: time to first token, output-token latency, key-value-cache efficiency, and inference-routing decisions.
Making autonomous agents operational
- Portable, containerized agents: O’Malley packages OpenClaw with its tools, skills, and MCP servers, allowing the same environment to move between local Podman, Kubernetes, and OpenShift. Her AI Engineer Europe demonstration also covers specialized sub-agents and the additional virtualization constraints of macOS.
- Layered credential isolation: Podman secrets or Kubernetes Secrets combine with OpenClaw secret references to keep API credentials separate from application configuration. Persistent volumes preserve agent state independently of disposable containers, supporting backup, recovery, and redeployment.
- A hardened operating system for agents: With tank-os, O’Malley extends containment to the host using Fedora bootc, rootless Podman, a largely read-only operating-system layer, separate persistent state, and transactional image-based updates. The project remains an experimental open-source prototype.
- Governed team-agent environments: Her workplace model starts with approved MCP servers, sanctioned authentication, shared tools, and team-specific skills; employees then personalize that baseline. OpenTelemetry and Jaeger provide optional runtime visibility.
O’Malley also organizes DevConf.US and teaches within Boston University’s Faculty of Computing and Data Sciences.
