Sally Ann O'Malley on OpenClaw in Containers

Sally Ann O'Malley, a Principal Software Engineer at Red Hat, presented "Lobster Trap: OpenClaw in Containers from Local to K8s and Back" at AI Engineer Europe. O'Malley discussed the advantages of running AI agents within containers, emphasizing how this approach enhances reproducibility, security, and portability.

From Local Setup to Kubernetes

O'Malley highlighted the common challenge of AI agent setups that work on a developer's local machine but are difficult to reproduce or deploy elsewhere. The "Lobster Trap" methodology aims to solve this by packaging OpenClaw agents and their dependencies into containers. This allows for a consistent and isolated environment, whether running locally via Podman or deploying to a Kubernetes cluster.

Key Benefits of containerized AI agents

O'Malley detailed the core advantages of this container-centric approach:

• Reproducible Environments: Using the same container image ensures that all agents operate in identical conditions, regardless of the underlying infrastructure.
• Secrets Isolation: Secrets, such as API keys, are managed securely within the container's environment, preventing them from being exposed to the host system.
• Portability Across Infrastructure: Containers can be easily moved between different environments, from a local machine to a virtual machine, or a Kubernetes cluster, with minimal changes.
• Volume-backed Persistence: Runtime state and data are persisted on volumes, ensuring that agent progress is maintained even if the container is restarted or moved.
• Security Boundaries: Containers provide a natural security boundary, isolating the agent's execution and preventing potential interference with the host system.

Secrets Management and Inference Providers

A crucial aspect of the presentation was secrets management. O'Malley explained that OpenClaw uses a SecretRef abstraction, allowing for different secret injection mechanisms depending on the environment. For local development with Podman, secrets are typically injected via environment variables. In Kubernetes, secrets are managed through Kubernetes Secrets. The system supports various inference providers, including OpenRouter, Anthropic, and Google, allowing users to choose their preferred models and manage API keys securely within the containerized setup.

Team-Wide Adoption and Reproducibility

O'Malley emphasized that this container-based approach is not just about individual productivity but also about fostering team collaboration and standardization. By providing a curated, reproducible baseline, new engineers can onboard faster and contribute effectively without getting bogged down in environment-specific setup issues. This also ensures that team standards for skills and model choices are shared, moving away from tribal knowledge.

The "Lobster Trap" Philosophy

The core idea, as O'Malley presented, is to "start local, curate, and reuse." This involves initially setting up agents locally, curating the necessary components and configurations, and then easily deploying and reusing this setup across different environments, including Kubernetes. This methodology transforms hard-won knowledge into a team asset, making the development and deployment of AI agents more efficient and reliable.

The presentation concluded with a demonstration of the OpenClaw installer, showcasing how easily agents can be deployed locally or to Kubernetes, highlighting the flexibility and power of containerization in the AI development workflow.