Part 2 – Running Your LLM Agent Safely: Hands-on with Docker Sandboxes

Docker Sandboxes exist because of a simple problem: the moment your LLM stops just answering questions and starts acting — calling tools, writing files, hitting APIs, making decisions on its own — you need somewhere safe to let it do that. An agent with unrestricted access to your systems is an agent with an unrestricted attack surface, and that’s exactly the gap this workshop closes.

This free, hands-on session is Part 2 of the Docker-sponsored webinar series. Over two hours, Docker’s Dan Ndombe will walk you through running LLM agents safely using Docker Sandboxes (SBX) — a practical, code-along workshop rather than a slide deck. No prior Docker experience is required.

What Are Docker Sandboxes — And Why Your LLM Agent Needs One

Once you understand how large language models work, the next logical question is where you run them. When an agent can execute code, browse the web, or touch your file system, “where” is no longer a deployment detail — it’s a security decision.

They give agents an isolated runtime with clearly defined boundaries: what they can access, what they can execute, and what they absolutely cannot touch. Instead of trusting an agent by default, you contain it by design. A single bad tool call, a malicious prompt injection, or an overly broad permission set stays boxed in, instead of turning into an incident.

What You’ll Learn in This Docker Sandboxes Workshop

This session is deliberately hands-on. By the end, you’ll have a repeatable pattern you can take straight back to your own projects. Specifically, you’ll:

• Spin up a sandboxed agent runtime using Docker Sandboxes (SBX)
• Assign a real agent a small, practical task and watch it execute inside the sandbox
• Explore how to tighten policies and permissions around agent behavior
• Walk away with a reusable pattern for deploying this approach in your own AI and agent prototypes

This isn’t abstract theory — it’s a live, practical demonstration of containment and control that you can replicate immediately on your own machine.

Who Should Attend

This webinar is built for a broad technical audience working anywhere near AI agents, including:

• Software engineers building or prototyping LLM-powered agents
• DevOps and platform engineers responsible for how AI workloads get deployed
• Security-minded developers who want practical, not theoretical, guardrails using tools like Docker Sandboxes
• Technical leads evaluating how to safely scale agentic AI initiatives

No prior Docker experience is required, so whether you’re brand new to containers or already comfortable with them, you’ll be able to follow along and apply what you learn.

As more teams move from experimenting with LLMs to deploying autonomous agents in production, isolating those agents is quickly becoming a non-negotiable part of the AI stack — not an afterthought bolted on after something goes wrong. According to NVIDIA’s technical guidance on sandboxing agentic workflows, controls like blocking unrestricted network access and file writes outside a defined workspace are considered baseline requirements for containing the risks introduced by autonomous, tool-calling agents. This workshop puts those principles into practice with Docker Sandboxes, so you leave with something you can implement, not just understand.

If you’re curious about the broader landscape of agentic AI beyond this session, Data Science Dojo’s ongoing coverage of LLM agents is a good next stop for deeper reading. Seats for this hands-on session are limited to keep the workshop practical and interactive — if your team is building, deploying, or even just experimenting with AI agents, learning to sandbox them properly is a direct, no-fluff place to start.