How AgentRidge works
AgentRidge behaves like a small, tireless red team packed into a desktop app. You hand it an engagement; it plans the work, runs the tools, interprets what comes back, and writes up what it found. This page is the bird's-eye schematic — each part links to a deeper page.
The system at a glance
Read it top to bottom: the operator supplies a mission and an authorization. Mission Control hands it to the Principal (the brain), which breaks the work into steps and delegates to Ridge Sub-Agents. Both the brain and the sub-agents reason using the Ridge inference engine running on your own machine. Sub-agents drive real security tooling, read the results, and stream confirmed findings back to Mission Control, which turns them into a live attack-surface map, a vulnerability register, and an exportable report.
The four building blocks
| Block | Role | Learn more |
|---|---|---|
| Agent Ridge Principal | The brain agent. Owns the plan, keeps the objective in mind, delegates, and decides when the engagement is done. | Principal & sub-agents |
| Ridge Sub-Agents | Specialists for recon, exploitation, and parsing. They generate commands and interpret output. | Principal & sub-agents |
| Ridge inference engine | The local reasoning core. Keeps engagement data on the operator's workstation. | Local-first design |
| Mission Control | The desktop UI: launch missions, watch progress live, and collect deliverables. | Workspace tour |
The mission loop
Every engagement follows the same autonomous loop, driven by the Principal:
The dotted arrows are the important part: AgentRidge loops back. A newly discovered host reopens recon; a partial exploit reopens enumeration. It keeps going until every in-scope lead is exhausted — not until the first "good enough" result. See the mission lifecycle for the full walkthrough.
Why it feels different
- It works the whole scope, autonomously, instead of answering one question at a time.
- It adapts — when a path is blocked (a firewall, a timeout, a filter), it documents the blocker, switches technique, and keeps attacking the rest of the surface.
- It proves impact with concrete evidence — exact commands, full request and response pairs, and the objects or credentials it reached.
- It reasons locally, so the sensitive parts of an engagement never have to leave the operator's machine.
Continue with the brain agent and its sub-agents →