agent red

the product

A red team that never clocks off.

agent red runs a crew of AI operators inside your own environment. They plan, probe, exploit and report around the clock — with you as closely involved as you want to be.

Start freeWhat it can do

where it runs

it runs on your infrastructure. never ours.

One command starts an agent on a machine you choose — a laptop, a VM in your VPC, a runner in CI, a box inside the client network. It dials out to a private tailnet only your org can see, opens no ports, and carries no AI keys. Your data and your targets never leave your environment.

an operation, end to end

Brief. Field. Intel. Debrief.

Each run is an operation — one discipline against a target you've proven you own, with a codename and a durable record. Reboot the box halfway through and it resumes where it left off.

  1. 01briefing

    scope it, prove you own it, sign the rules of engagement

  2. 02in the field

    operators work it — watch, steer, or walk away

  3. 03intel

    scored findings land as they're proven, worst first

  4. 04debrief

    fix it, then send them back in to retest

how involved are you?

Assist, take over, or set and forget.

Switch between them whenever you like, on any operator. The same operation can start hands-on and end unattended.

01assist

You drive. It rides along.

Work in the terminal yourself and hand the operator the tedious parts — enumeration, payload variants, the write-up. Every risky step waits for your yes.

02take over

Grab the wheel mid-run.

Open the live terminal on any operator, from the browser or over SSH, and take control of its session. Hand it back when you're done and it carries on from there.

03set & forget

Brief it and walk away.

Give it the objective and the rules, then leave. Operators work around the clock, survive restarts, and only call you when they've proven something or need sign-off.

the crew

A handler and a team of operators.

The handler reads your brief, plans the operation and places work across your agents. Each operator is a specialist working its own terminal. They leave each other notes on a shared board, read each other's transcripts, and flag what's missing from their toolkit so it gets built.

memory
a durable journal of what it saw and why it decided
restarts
the operation resumes, it doesn't start over
models
Claude, GPT, Gemini and more — picked per job
skills
a shared library of playbooks every operator draws on

you set the blast radius

Wall it off as hard as you like.

Pick how contained the agent is from your host, and how loud it is on the wire. Stealthy keeps outbound traffic proxied so it won't trip your own perimeter alerts; noisy goes direct for fast external testing.

  1. weak

    a plain process on the host — fastest, for disposable lab boxes

  2. medium

    a standard container — full tooling, raw-socket scans included

  3. strong

    a gVisor sandbox — a hard syscall boundary around the agent

  4. strongest

    a burner box — your own host is never in the blast radius

tradecraft

Nineteen disciplines. Start with one.

You don't plan a six-month program. Start at surface monitoring today — one click — and each operation opens the next, all the way to a goal-seeking adversary that tests whether your SOC notices. Every discipline is open to everyone; none sits behind a plan. See the guided plan

start here

then

then

then

then

deploy anywhere

Wherever you can run a container.

runtimes
docker · podman · orbstack · nix
platforms
linux · macos · windows (wsl2) — x86_64 and arm64
in ci
runs on every push; report-only or block the merge
inside the network
stealthy egress for monitored client LANs
outside the perimeter
a black-box attacker's view
setup
one command, keys baked in, nothing left in shell history

only test what you own

One command. A crew on shift. Intel by morning.

Start free and point it at something you own.

Start freeSee the pricing