Blog
Commt blog
23 August 2026
How to limit what an AI agent can reach
An AI agent's reach is not one setting. It is four separate boundaries: what it can call on the network, which tools and connectors it can invoke, which knowledge it can retrieve, and what gets kept afterward. This post treats each as an engineering problem and names the tradeoffs instead of hiding them, for anyone building the boundary rather than shopping for a claim about one.
Read more23 August 2026
What is a private AI agent?
Most explanations of "private AI agent" stop at self-hosting or lean on a model provider's no-training promise, and both miss what actually makes an agent private: four boundaries you can check yourself, covering what it knows, what it can do, what it can reach over the network, and what gets kept afterward. This post breaks down each boundary, explains why self-hosting and no-training claims are necessary but not sufficient on their own, and lists the specific questions worth asking a vendor before trusting an agent with real data.
Read more