Arrakis

open source

Open-source self-hostable sandboxing service for secure AI agent code execution and GUI automation

Updated 2026-03-17

Tool EcosystemsFree· Open source; self-hosted on your own infrastructureVisit site Docs

Arrakis is a self-hosted sandboxing platform that enables AI agents to securely execute code and interact with graphical interfaces in isolated MicroVM environments. Built by an infrastructure veteran from Replit and Google, it provides snapshotting and backtracking capabilities, integrates natively with Claude via MCP, and ships with Python SDK and MCP server support out of the box.

Use Cases

·Enable AI agents to execute arbitrary code safely without contaminating host systems
·Allow Claude and other LLM agents to interact with browser-based UIs and GUI applications
·Build autonomous agents that can debug, iterate, and recover from execution failures via snapshotting
·Create AI-native tools like document editors or spreadsheets that agents control end-to-end
·Run multi-step agent workflows that require both code execution and visual interaction

Key Points

·First-class snapshotting and backtracking enables agents to recover from failures and iterate autonomously without human intervention
·MicroVM-based isolation provides strong security boundaries suitable for running untrusted or complex agent code safely
·Native MCP integration allows Claude and compatible models to control sandboxes directly, including GUI interaction
·Self-hosted architecture ensures data privacy, cost predictability, and infrastructure control, critical for solo AI businesses
·Python SDK and Docker customization enable bespoke agent workflows and sandbox configurations

Arrakis

Use Cases

Key Points

Tags