NanoClaw partners with Docker to deploy AI agents in secure sandboxes after viral six-week rise

NanoClaw's Meteoric Rise From Weekend Project to Docker Partnership

What started as a weekend coding marathon has transformed into a partnership between NanoClaw and Docker that could reshape how enterprises deploy AI agents. Gavriel Cohen built NanoClaw in approximately 48 hours about six weeks ago, introducing it on Hacker News as a tiny, open-source AI agent platform designed as a secure alternative to OpenClaw . The project exploded after famed AI researcher Andrej Karpathy praised it on X, leading to 22,000 stars on GitHub, 4,600 forks, and over 50 contributors within weeks .

On Friday, Cohen announced a formal partnership with Docker to integrate Docker Sandboxes into NanoClaw, allowing the open-source AI agent platform to run in isolated micro VMs with a single command 2

2

. Docker, which essentially invented secure container technology and serves nearly 80,000 enterprise customers, provides the infrastructure layer that addresses one of the biggest obstacles to enterprise agent deployment: how to give AI agents operational freedom without exposing host systems to security vulnerabilities .

Why Security and Isolation of AI Agents Matters for Enterprises

The partnership tackles a fundamental tension in enterprise adoption of AI agents. These systems need broad access to perform useful work—installing packages, modifying files, accessing credentials, and connecting to external systems. But each capability increase raises containment stakes. "Every organization wants to put AI agents to work, but the barrier is control: what those agents can access, where they can connect, and what they can change," said Docker president Mark Cavage 2

2

.

Cohen's motivation for building NanoClaw stemmed directly from security concerns with OpenClaw. While using OpenClaw for his AI marketing startup, he discovered the agent had downloaded all his WhatsApp messages—both work and personal—and stored them in plain, unencrypted text on his computer . OpenClaw has been widely criticized as a "security nightmare" due to how it accesses memory and account permissions, with a codebase sprawling across 800,000 lines of code that's difficult to audit .

In contrast, NanoClaw was built with just 500 lines of code initially, now supported by fewer than 4,000 lines, making it far easier to examine for errors and security issues 2

2

. "Application-level permission checks don't offer sufficient protection. What is required is OS-enforced isolation: each agent in its own safe environment, with its own filesystem and session history, invisible to every other agent running alongside it," NanoClaw's team noted 2

2

.

How Docker Sandboxes Create Safer AI Agents Through Containerization

Docker Sandboxes represent a shift from standard containerization to MicroVM-based isolation. While traditional containers are isolated processes on a shared kernel, micro VMs have their own dedicated kernel and hardware space, creating a stronger security boundary 3

3

. "With Docker Sandboxes, that boundary is now two layers deep," Gavriel Cohen explained. "Each agent runs in its own container (can't see other agents' data), and all containers run inside a micro VM (can't touch your host machine)" 3

3

.

This architecture addresses the reality that AI agents fundamentally break assumptions built into conventional infrastructure. "Agents break effectively every model we've ever known," Cavage said. "Containers assume immutability, but agents break that on the very first call. The first thing they want to do is install packages, modify files, spin up processes, spin up databases—they want full mutability and a full machine to run in" 4

4

. According to NanoClaw, if an agent tries to escape by exploiting a vulnerability, it remains contained within the disposable isolation zone 2

2

.

The integration means enterprises can experiment with NanoClaw without loading agent software directly onto host systems—a risk that can lead to accidental deletion, damage, and prompt injection attacks 2

2

. Docker Sandboxes are currently supported on macOS (Apple Silicon) and Windows (x86), with Linux support expected in a few weeks 3

3

.

From AI Marketing Startup to Full-Time Focus on Enterprise Agent Deployment

Cohen's journey to this partnership began when he launched an AI marketing startup with his brother Lazer Cohen, offering services like market research and go-to-market analysis through AI agents. The agency was on track to hit $1 million in annual recurring revenue before Cohen closed it down about a week ago to focus full-time on NanoClaw and launch a company around it called NanoCo .

The partnership with Docker emerged organically when Oleg Selajev, a Docker developer, saw the buzz around NanoClaw and modified it to replace Apple's container technology with Docker Sandboxes . Cohen integrated the changes into the main NanoClaw project, recognizing it was "no longer my own personal agent that I'm running on my Mac Mini" but a tool serving a broader community .

For technical leaders watching this space, the partnership signals a shift from AI agents as novelty to production-ready infrastructure. "Once you get there, then the developers all of a sudden go from effectively babysitting the agent to just letting it run for minutes or hours or longer at a time," Cavage said. "That's the huge productivity unlock" 3

3

. The challenge ahead involves reconciling the deterministic nature of computers with the non-deterministic nature of AI models—something Cavage admits "is not a solved problem and is something that will occupy the industry for a while" 3

3

.