A developer released stoat, a Go-based open-source tool giving coding agents safe VM access via an MCP server instead of shared-kernel containers. Motivated by two incidents of destructive shell commands on the host machine, stoat exposes exec, files, packages, ports, services, logs, and snapshots inside per-guest VMs. It matters to practitioners running agents autonomously who need stronger isolation than containers provide.
stoat is an agent-native tool that lets LLM agents spin up and manage sandboxed QEMU virtual machines for code execution. It solves the gap between Docker (lacking systemd/kernel modules) and GUI-driven VirtualBox/VMware, so developers running Claude Code or similar can sandbox agent work without approving every bash command or exposing their host machine.
stoat is an agent-native tool that lets LLM agents spin up and manage sandboxed QEMU virtual machines for code execution. It solves the gap between Docker (lacking systemd/kernel modules) and GUI-driven VirtualBox/VMware, so developers running Claude Code or similar can sandbox agent work without approving every bash command or exposing their host machine.
Krowoc is an open-source agent workspace that gives AI agents their own micro-VM with shell, filesystem, and headless Chrome, allowing long-running tasks and scheduled jobs to continue without keeping a laptop open. It supports multiple model providers and emphasizes security by isolating agents from shared containers.
Krowoc is an open-source agent workspace that gives AI agents their own micro-VM with shell, filesystem, and headless Chrome, allowing long-running tasks and scheduled jobs to continue without keeping a laptop open. It supports multiple model providers and emphasizes security by isolating agents from shared containers.
A developer is exploring an architecture for running AI agents inside isolated, virtualized environments using Firecracker microVMs, aiming to simplify the developer experience by abstracting away low-level configuration. The proposed stack is AI Agent → Sandbox API/SDK → Firecracker microVM → isolated Linux filesystem, with the agent running outside the VM while code execution happens inside.
A developer is exploring an architecture for running AI agents inside isolated, virtualized environments using Firecracker microVMs, aiming to simplify the developer experience by abstracting away low-level configuration. The proposed stack is AI Agent → Sandbox API/SDK → Firecracker microVM → isolated Linux filesystem, with the agent running outside the VM while code execution happens inside.
Isolade is an open-source, local-first workbench that runs coding agents in isolated microVMs, combining agent management with microVM isolation. It provides each agent its own microVM with subsecond provisioning and domain-scoped secret substitution, so the VM only sees placeholder values, addressing the need for secure, isolated agent execution. It targets developers who want to run coding agents locally without exposing secrets.
Isolade is an open-source, local-first workbench that runs coding agents in isolated microVMs, combining agent management with microVM isolation. It provides each agent its own microVM with subsecond provisioning and domain-scoped secret substitution, so the VM only sees placeholder values, addressing the need for secure, isolated agent execution. It targets developers who want to run coding agents locally without exposing secrets.
This project provides each AI agent session with an isolated Firecracker microVM, including isolated networking, CPU/memory limits, egress policies, persistent workspace, streaming command execution, MCP integration, and multi-tenant isolation. It aims to make agent execution safe, fast, and production-grade, solving the problem of secure and scalable agent sandboxing.
This project provides each AI agent session with an isolated Firecracker microVM, including isolated networking, CPU/memory limits, egress policies, persistent workspace, streaming command execution, MCP integration, and multi-tenant isolation. It aims to make agent execution safe, fast, and production-grade, solving the problem of secure and scalable agent sandboxing.