Give every agent its own space.
Separate execution environments designed to keep agent activity away from sensitive host resources.
A home for your AI agents. Run them in isolated environments, see what they're doing, and keep your machine yours.
One clean workspace for running and observing your coding agents. No mystery. No mess.
↗ CONCEPT PREVIEW — THE REAL DESKTOP APP IS IN DEVELOPMENT
Separate execution environments designed to keep agent activity away from sensitive host resources.
Follow execution logs, file changes and security events in one focused interface.
Define what an agent can reach, review its work and stop a session when needed.
Because giving an AI your terminal shouldn't mean giving it the keys to everything.
AgentSandbox is being designed as a free, open-source tool for people who build with AI. Transparent development, honest security boundaries and a community-first approach.
const agentsandbox = {
access: 'least-privilege',
visibility: 'always',
source: 'open',
price: 0
};
We're starting with Windows, one reliable isolation backend and a focused desktop experience.
Windows desktop shell, mascot, UI system and agent launcher.
Verified sandbox execution, activity logs and permission controls.
Additional integrations, change review and richer workflows.
AgentSandbox is taking shape. Check back as the project moves towards its first public release.
Back to the top ↗