Northbeams: An AI security analyst for the Shadow AI your team uses

Hi Product Hunt 👋 I'm Joe, founder of Northbeams. Your team adopted AI faster than anyone approved. In a typical 50-person company we find 27 active AI tools and 143 sensitive prompts sent to public models every month. Most of it never touched IT. You are already accountable for AI you cannot see. Northbeams is the AI System of Record. One platform watches four surfaces: the browser, the desktop, the CLI, and the MCP traffic between your coding agents and their tools. It catches AI activity the day it shows up, classifies it on-device, and writes it to one audit log. The part I am most excited to show you is new. We rebuilt the dashboard into an AI security analyst. It reads the raw signal, writes you a plain-English brief, and hands you a decision queue: 'Cursor just called an MCP tool that can write to your repo. Allow, warn, or block?' You approve. It acts. Every action is reversible and logged. No LLM ever sits in the allow/block path, so enforcement stays deterministic. For the engineers here: the MCP Gateway sits in-path between your agents (Claude Desktop, Cursor, Claude Code) and their MCP servers. It classifies the actual arguments on-device and enforces allow, warn, or block per individual tool. The values never leave the laptop. No proxy, no MITM cert, no network change. You can see all of it without signing up. The demo is a full tour of the live product: https://monitor.northbeams.com/demo Free to start. Published pricing, no sales call to see a number. Install on Mac, Windows, and the browser, and you will have your first detections by Monday. I will be in the comments all day. Tell me what your team is running that you wish you could see. What would make this an instant install for you? Joe McCormack Founder, Northbeams