TL;DR: we created a minimized OpenClaw container image that contains 99% less vulnerabilities than the original one. The functionality is exactly the same. The image is regularly updated and is released for free. You can start using it now.
docker pull us-docker.pkg.dev/prod-375107/minimus-public/openclaw:latest
OpenClaw lets you have Claude, GPT, or other AI models available on your messaging apps (WhatsApp, Telegram, Discord, Slack) without sending everything through third-party services. This is great if you want to control your data and own the infrastructure. However, it also means giving an application deep access to your accounts. With that level of access, OpenClaw can take action on your behalf, acting like a true personal assistant with capabilities far beyond that of today’s AI chat services.
The power of OpenClaw comes from the connections it has to your accounts and system. To function fully, OpenClaw requires access to:
With this level of access to your life, security needs to be at the forefront of any OpenClaw installation to ensure that attackers can’t compromise your system, exfiltrate your sensitive data, or prompt OpenClaw into malicious action.
One of the most common ways to run OpenClaw is in a container, and there are several popular distributions for doing so. Our Minimus Labs research team ran CVE scans on these distributions and the results were…not great.
Important: The "official" OpenClaw image from GitHub Container Registry has 2,062 known CVEs, even more than Docker Hub versions!
Plot twist: alpine/openclaw isn't even Alpine Linux - it's Debian 12 with 1,156 vulnerabilities! You can verify this yourself: docker run --rm alpine/openclaw cat /etc/os-release
Note: These are snapshot numbers from today’s scan. CVE counts change as new vulnerabilities are discovered and patched.
The Minimus OpenClaw image has many improvements over the official image, including:
pnpm and tar, normal for any Node.js app, and easily updated with our regular builds.
To illustrate the security gap, we scanned several widely used OpenClaw container images and compared them directly with the Minimus image:
Scan Results:
├── 2,062 TOTAL vulnerabilities ❌❌❌
├── 7 CRITICAL vulnerabilities ❌
├── 233 HIGH vulnerabilities ❌
├── 928 MEDIUM vulnerabilities ❌
├── 890 LOW vulnerabilities ❌
└── 413 OS packages to track ❌
Base OS: Debian 12.13 (bookworm)
Image Size: 4.66 GB
Security Posture: EXTREMELY POOR
Recommendation: AVOID - Being "official" does NOT mean secure
Critical Issues:
├── CVE-2025-7458: SQLite integer overflow (NO FIX)
├── CVE-2023-45853: zlib heap buffer overflow (NO FIX)
├── CVE-2023-6879: libaom heap overflow (NO FIX)
├── CVE-2023-5841: OpenEXR heap overflow (NO FIX)
├── CVE-2025-48384/48385: Git arbitrary code execution (NO FIX)
└── CVE-2026-0861: glibc integer overflow (NO FIX)
Scan Results:
├── 2,054 TOTAL vulnerabilities ❌❌❌
├── 7 CRITICAL vulnerabilities ❌
├── 242 HIGH vulnerabilities ❌
├── 907 MEDIUM vulnerabilities ❌
├── 888 LOW vulnerabilities ❌
└── 413 OS packages to track ❌
Security Posture: EXTREMELY POOR
Recommendation: AVOID - highest vulnerability count
Scan Results:
├── 1,156 TOTAL vulnerabilities ❌
├── 7 CRITICAL vulnerabilities ❌
├── 242 HIGH vulnerabilities ❌
├── 907 MEDIUM vulnerabilities ❌
└── 413 OS packages to track ❌
Security Posture: POOR
Recommendation: NOT for production use
Scan Results:
├── 0 CRITICAL vulnerabilities ✅
├── 4 HIGH vulnerabilities ⚠️ (npm packages)
├── 3 MEDIUM vulnerabilities ⚠️ (npm packages)
└── ~50 packages to track ✅
Security Posture: EXCELLENT
Recommendation: Production-ready
Minimus OpenClaw image:
Look, if you're running OpenClaw, you probably:
The problem is most "self-hosted" guides just say "docker pull whatever" and call it a day. That's fine until you realize your default setup contains more than 2k known vulnerabilities and you just gave it the keys to your WhatsApp.
Using Minimus OpenClaw Image provides a more secure starting point. Without changing OpenClaw’s functionality, you get:
It's self-hosting done right, not just self-hosting done quickly.
Note: Please be aware that there are additional security implications to consider when deploying any OpenClaw agent in a containerized environment using Docker default configurations from the Upstream OpenClaw repo. We suggest our readers go over all configurations and adjust them accordingly.
Running your own AI assistant shouldn't mean compromising on security. Try the Minimus OpenClaw image to get:
Getting started with Minimus' OpenClaw image is simple:
1. Clone the OpenClaw GitHub repo.
git clone https://github.com/openclaw/openclaw.git
cd openclaw
2. Switch to the Minimus OpenClaw image.
echo "FROM us-docker.pkg.dev/prod-375107/minimus-public/openclaw:latest" > Dockerfile
3. Continue with either the Docker option, or the Podman option. Below, we show the Podman option.
./setup-podman.sh
Compose manual flow will also work seamlessly.
Docker Hub shows compressed download sizes (~1.6 GB), but actual on-disk usage is what matters. The Minimus image uses ~2.23 GB uncompressed vs ~4.0 GB for alpine/1panel - slightly smaller with vastly better security. The real win is 99% fewer vulnerabilities and minimal attack surface.
All 11 are in npm packages (pnpm, tar), not the OS. These are:
Yes! Our images are multi-arch: linux/amd64 and linux/arm64
Yes! OpenClaw is MIT licensed. Our images follow the same license. More information about licensing and copyrights.
.svg)
.svg)
.svg)
.svg)
