Skip to main content

Introducing Clawrium: An aquarium for *claws

· 3 min read
Devashish Meena
Devashish Meena

I installed my first OpenClaw on an old Ubuntu box. Smooth. Then I dug up a Raspberry Pi and put ZeroClaw on it. A week later I was spending more time on SSH gymnastics than actually using the agents — upgrading one, swapping a model on another, debugging logs on a third, rotating a token on the fourth. The agents were fine. The orchestration around them was the problem.

That's when it clicked: this is a standardization and orchestration problem, and it looks a lot like what Kubernetes solved for containers. Running one agent is easy. Running five across your network while juggling SSH keys, model configs, personas, and channel integrations is where it falls apart.

So I built Clawrium — an aquarium for *claws. A single CLI (clm) that treats your AI agents as a fleet.

What clm does today​

  • Deploy multiple agent types (OpenClaw, ZeroClaw, Hermes) across hosts on your local network or cloud
  • Normalized configuration, secrets, and integrations across every supported agent type
  • Swap models, rotate secrets, update personas without SSH-ing anywhere
  • Install reusable skills onto agents from a shared registry
  • One status pane (clm ps) for the whole fleet

The aspiration is broader — backups, cost guardrails, central log aggregation, drift remediation, agent cloning — and I'm building it in the open as the use cases sharpen. See the docs for the full shape of what's wired up today.

Why now​

The agents I talk to teams about are no longer hobby projects. They're solving real work — but every team I've spoken to manages more than one, and once you cross that threshold, the operational burden eats the productivity gain. One org I talked to runs a separate OpenClaw per engineering team. The 90% of configuration that's shared (Linear, Slack, Confluence) is duplicated and managed differently in each instance. Clawrium pulls those agents under one umbrella.

In the pets-vs-cattle framing, Clawrium agents are still pets. They're just pets that are well-trained, on a leash, and play well with each other.

What I'm building with it​

Two real use cases drive the roadmap:

  1. A project assistant for Clawrium itself — an agent on my local network that I talk to over Discord. It files issues, gives status updates, captures notes. The bot's name is Maurice; if you join the Discord, be nice to it.
  2. Team assistants at work — one agent per team, tuned to their domain and ownership. Reduces the logistics tax on humans.

Try it​

If you want the full picture, the architecture page covers how the layers fit together, and the installation page is the canonical step-by-step.

The 30-second version:

uv tool install clawrium
clm host init <ip> --user <user>
clm agent install --type openclaw --host <alias>