Devvm.dev: Isolated Virtual Machines for Projects and AI Agents

Devvm.dev introduces a streamlined approach to running development projects and coding agents inside isolated sandboxes. The tool operates seamlessly on both macOS and Linux, utilizing robust virtualization backends to keep host environments secure.

• Start a fully isolated virtual machine for your project with a single command.
• Native support for coding agents like Claude Code, Codex, and opencode.
• Automatic port forwarding and direct integration with VS Code via Remote-SSH.
• Robust protection of your host machine and git repositories from rogue agents or supply chain risks.

How Devvm.dev Works

The installation script sets up requirements like Lima and QEMU with explicit user prompts. When run inside a project folder, devvm automatically detects required toolchains or configures the environment based on devcontainer.json settings.

Network tiers and a local proxy ensure that sensitive credentials, such as Claude Code authentication tokens, never leave your host machine. The virtual environment interacts with placeholders while the proxy injects actual tokens into outbound requests.

Security and Sandbox Isolation

Developers face increasing threats from supply chain vulnerabilities, leaked secrets, and unpredictable behavior from autonomous coding agents. Running these tools within a strict sandbox limits the blast radius of any potential security incident.

On macOS, devvm locks local git hooks and configurations to prevent virtual machines from altering repository settings on the host. Additionally, the --clone flag allows agents to work on an isolated repository copy, preventing direct modifications to your working tree.

Architecture and Performance

The first virtual machine builds a shared base image, allowing subsequent environments to boot up in mere seconds. Toolchains are compiled once and shared across VMs in read-only mode to conserve storage and optimize resource usage.

Developers can further customize their environments by creating bespoke setup layers, adding terminal utilities like fzf and zoxide, or initializing virtual machines directly from standard container images.

Significance for Developers

Tools like Devvm.dev lower the barrier to securely adopting AI coding assistants in everyday software development. By containing agent activities within secure boundaries, developers can leverage automation without compromising host system integrity.