5 comments

[ 2.7 ms ] story [ 20.9 ms ] thread
Nomos is trying to make that boundary explicit at execution time, but I’m curious how others are handling it today.

For example: - do you restrict tools at the framework level? - rely on sandboxing? - or allow broad access and rely on monitoring/audit?

[flagged]
Yeah, I am also thinking to make it full-pledged MCP gateway, so we don't need to attach multiple MCP servers directly to agent, rather all go through single gateway with policy enforcement.