OpenClaw, Without the Ops

OpenClaw is an open-source AI agent runtime. Launched in November 2025, it hit 100,000 GitHub stars in its first ten weeks and surpassed React to become the most-starred software project on GitHub. That adoption happened because it works: persistent agents that connect to your messaging platforms, remember context, and get better as you correct them.

It's not a chatbot. It's an agent runtime: a persistent process that connects to your tools and your LLM provider of choice. It executes multi-step tasks autonomously. All data is stored as plain files on disk, not in a proprietary database.

Every vessel ships with built-in tools that work immediately. No plugins to install, no extensions to configure. Your agent can read and write files, browse the web, remember context across conversations, and run scheduled tasks from the moment it boots.

Your agent connects to the platforms you already use. All channels run simultaneously with shared memory, so the agent has the same context whether you message it on WhatsApp or Slack.

You probably don't care about MIT licenses. But you do care about what happens when a vendor shuts down, raises prices, or changes terms. Open source means the code running your agent is publicly inspectable by anyone. It's maintained by a large open-source community under the MIT license, and can't be pulled out from under you.

Compare that to proprietary agent platforms where you can't see what the code does with your data, can't verify security claims, and can't leave without losing everything. OpenClaw's data is plain files. If you ever want to move, you copy a folder.

OpenClaw is powerful enough that developers build complex agent systems with it. But power comes with operational complexity. Setting it up properly requires Docker, VM management, networking, authentication configuration, SSL certificates, and ongoing maintenance. Most knowledge workers shouldn't need to know what any of those words mean.

And when the deployment is done poorly, the consequences are real. CrowdStrike published a detailed security analysis of OpenClaw deployments. The findings were sobering.

This isn't OpenClaw's fault. It's what happens when a powerful developer tool gets deployed without security expertise. The same thing happened with Redis, MongoDB, and Elasticsearch before it. Powerful tools need proper deployment.

The people who benefit most from AI agents are professionals whose time is too valuable to spend on DevOps. Lawyers reviewing contracts. Consultants building deliverables. Marketing strategists running campaigns. They need the agent, not the ops.

NVIDIA recognized the same gap. Their open-source NemoClaw project adds a kernel-level sandbox layer to self-hosted OpenClaw deployments: deny-by-default network policies, filesystem restrictions, and process isolation. Vessel takes a different approach. Rather than sandboxing OpenClaw on a shared machine, each vessel runs on its own dedicated virtual machine. A hardware boundary, not a process boundary.

Every vessel is a dedicated virtual machine on Google Cloud with OpenClaw pre-configured and running. No Docker setup. No networking. No security configuration. You provision a vessel from the dashboard and your agent is running in minutes.

Behind the scenes, each vessel runs a sidecar process that monitors your agent's health, restarts it if it crashes, applies daily updates, tracks resource usage, and alerts on failures. You don't see any of this. Your agent just works.

Your data stays on your dedicated VM. No public IPs. No SSH. Inbound access only through encrypted Cloudflare Tunnels. The infrastructure isolation that matters for professionals handling sensitive work.

Vessel isn't for developers who want to tinker with agent infrastructure. It's for professionals who need AI agents to get their work done: lawyers who need contract review, consultants who need research synthesis, marketers who need campaign execution, financial advisors who need compliance monitoring. Knowledge workers whose time is spent on expertise, not on keeping servers running.

If you can use Slack, you can use a vessel. The agent handles execution. You handle judgment.

AI agents process client data, draft confidential documents, and access your accounts. An agent with the wrong security configuration isn't just a technical problem. It's a liability. 135,000 OpenClaw instances were found exposed to the internet with no authentication. That's not a hypothetical risk.

Every vessel runs on a dedicated virtual machine with its own kernel, no public IP, no open ports, and no access from other tenants. Inbound traffic goes through encrypted Cloudflare Tunnels. The infrastructure is hardened before your agent ever boots. Health monitoring, automatic restarts, daily updates, security patches: all handled, all the time.

You don't need to think about whether your agent is secure. You don't need to wonder if it crashed overnight, or if an update broke something, or if someone found an open port. That's what we do. See how provisioning works, then focus on the work the agent is doing for you.