Arrakis: How To Build An AI Sandbox From Scratch
summary
Arrakis is an open source microVM sandbox that claims secure, fast code execution with boots in less than 7 seconds and a PR aiming for under one second, uses cloud-hypervisor + KVM, exposes a REST API plus a Python SDK and Model Context Protocol (MCP) server, provides VNC/browser access with host-managed port forwarding and per-VM overlayfs for immutable base + writable layer, implements snapshot and restore by pausing the VM, dumping guest memory, persisting the RW layer, and resuming so agents can backtrack and retry multi-step workflows.
AI-written summary · about
chapters (17)
- 0:00AI sandboxing and Arrakis intro
- 1:05AI agents need secure sandboxes
- 2:47Arrakis sandbox features
- 3:30MicroVMs for fast secure execution
- 8:10Linux sandboxing and containers
- 13:12Container security limitations
- 14:59Virtualization and microVMs explained
- 18:32What makes microVMs different
- 21:02Why Arrakis chose microVMs
- 23:54File system isolation in sandboxes
- 27:10Networking for AI sandboxes
- 29:38Customizing sandboxes with Docker
- 31:22Code execution in sandboxes
- 32:45Snapshotting for reliability
- 35:57Arrakis Python SDK
- 37:01Live demo with Claude Desktop
- 39:11Arrakis future work
related talks
12mDocker Sandboxes for AI
29mSandbox for Agentic Application With Cloud Native Stack
24mContained Intelligence: Secure Sandboxing for Gen AI Code
25mPlease Sandbox your AI Agents
25mFrom Cloud To AI: How Kata Containers 4.0 Reinvents the Sandbox for the Agent Era
38mWhy, and how you need to sandbox AI-Generated Code?
28mSecure AI Agent Sandboxing With OpenStack Zun and Kata Container
34mZero Secrets: Building Secure Infrastructure for AI Agents