Sandbox every integration. Keep it fast. Isolate every interaction. Contain anything that misbehaves.
Enclaves
Connecting agents to your systems means running integration code, sometimes code you didn't write. On Metorial, that code never runs loose. Every integration runs inside an enclave: a fast, secure sandbox that keeps it isolated and monitored. The convenience of connecting tools never comes at the cost of security.
All integration code runs inside an isolated enclave. It's always contained and never running directly on shared infrastructure.
Enclaves are built to be lightweight and quick as well as strongly isolated. Strong security doesn't slow down the work agents are doing.
It isn't one sandbox per customer. Every user, connection, and tool call gets its own enclave. The blast radius of anything going wrong stays tiny.
How enclaves protect you
Enclaves are the runtime boundary that the rest of Metorial's protections build on.
How the sandboxes work
Sandbox every integration. Keep it fast. Isolate every interaction. Contain anything that misbehaves.
Answers to common questions about enclaves and how they fit into governed AI agent infrastructure.