Security
Isolation, credentials, autonomy and what Modem can see.
Isolation
- One micro-VM per account, with its own kernel, its own disk and its own access token. Nothing is shared between accounts.
- Unprivileged user. Agents run as
dev, never as root, with no access to the host. - No path outward. A computer holds no credentials for Modem's infrastructure, so nothing on it can create machines or reach another account.
- No public ports. The only exposed surface is the daemon, behind your token.
Your credentials
Agent and GitHub logins are performed by you, inside your computer, and written to its disk. Modem does not collect, store or proxy them. The same applies to anything you push: files stay on your volume.
Why headless agents run without prompts
A permission prompt only means something when a human can answer it. In a job, nobody can: so the agent is granted autonomy within the VM, and the guarantees move to the walls: an isolated machine, a hard timeout, no reach into your laptop, and Git as the review boundary for code.
Give a computer the access the work needs, and no more. It is a workbench, not the place for production database credentials.
Access to your computer
Every request carries a token created at provisioning and stored in ~/.modem/config.json on your machines. Without it, the computer answers 401, including from us. Losing a laptop means rotating that token; ask and we will.
Data location
Computers and the control plane run in Paris (EU). Disks are encrypted at rest, and traffic between your terminal and your computer is TLS end to end.
Reporting something
Found a hole? Write to dom@fabriquestudio.fr. Modem is an alpha built by one person; reports get answered personally, and quickly.