← sources

Blog about my Agent's architecture

source: reddit (reddit:AI_Agents)published 2026-10-11 20:34ingested 2026-10-12 02:35

retained text

I wrote this blog covering the isolation model and architecture of my AI agent product, Nellie. I'm currently serving about 20 households on this architecture and so far so good.

The foundations are having no inbound connections from the internet on the box hosting the agents, kernel-level security and isolation controls between accounts and treating the model itself as untrusted input and designing its harness and toolset around that rather than trying to rely on prompting to control its misbehaviors.

Looking for thoughts and feedback. Thanks.

document fetched 2026-10-12 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1x3irn1/blog_about_my_agents_architecture