About Semantic

Semantic builds portable sandboxes with enforceable permissions, so untrusted AI agents can run safely on any computer.

Why

Agents write code, run commands and push branches. Isolating them in a VM stops them from breaking the machine, but not from misusing what they are given. We give every agent its own microVM and a policy you control, enforced outside the agent.

What we ship

• A sandbox runtime: one microVM per agent, with snapshot, pause and resume of memory, disk and processes.
• A permission gateway for Files, Git and SSH, with credentials kept on the host.
• A tamper-evident agent log and a denials record.
• A managed control plane, with on-prem for enterprise.

Team

We have backgrounds in AI research and on-prem compute platforms, based in Cambridge, MA.