01
FOUNDATIONS / 2026-07-26
What is an AI Agent Sandbox?
From one-off code execution to a stateful, policy-aware, observable runtime for agents.
RESEARCH ARCHIVE / OPEN NOTES
Sandbox Research
Tracking versions, methods, sources, and limitations. Conclusions evolve; evidence remains.
02
ARCHITECTURE / 2026-07-26
Container, gVisor, Kata, and MicroVM
Comparing isolation models for agent workloads across security, startup time, compatibility, and operations.
03
LANDSCAPE / 2026-07-26
Open-source Agent Sandboxes in 2026
The architecture, boundaries, and use cases of E2B, OpenSandbox, CubeSandbox, Microsandbox, and OpenShell.