Audnaudn
CLI · Claude Code forkKYC-gated

audncode

A terminal-based coding harness for offensive cybersecurity. A fork of Claude Code, built on OpenClaude, shipping unlimited tokens of the uncensored Pingu Unchained models — including the new KONG and GODZILLA builds.

Not a tool your orchestrator calls. An autonomous agent that plans, recons, executes and retests on its own loop, on targets you are authorized to test.

Download audncode Start a free trial

Requires an active penclaw.ai subscription · GovID + selfie verification · limited seats

01 — Install

Three steps, any machine.

Download the installer for your OS, run it, then launch audncode. First launch opens the browser to sign you in against your PenClaw account and your KYC status — after that the command is available everywhere in your shell.

  1. 01

    Download

    Pick your platform below. Signed installers for macOS, Windows, Debian/Ubuntu, plus a distro-agnostic bundle for everything else.

  2. 02

    Run the installer

    Double-click, or use the one-liner on the card. It installs the audncode binary and its alias openclaude onto your PATH.

  3. 03

    Launch & sign in

    Run audncode in any project directory. It signs you in, verifies your subscription and KYC, and pulls your PenClaw marketplace tools automatically.

macOS
.pkg · 13 MB
Download

macOS 11+ · Apple Silicon or Intel · signed & notarized

sudo installer -pkg AudnCode-Installer.pkg -target /
Ubuntu / Debian
.deb · 59 MB
Download

amd64 or arm64 · apt-managed upgrades

sudo apt install ./audncode_1.0.3_all.deb
Windows
.msi · 62 MB
Download

Windows 10/11 · x64 · adds audncode to PATH

msiexec /i AudnCode-Setup.msi
Any Linux
.run · 94 MB
Download

Distro-agnostic self-extracting bundle — Fedora, Arch, RHEL, Alpine

chmod +x AudnCode-Linux-AnyDistro.run && ./AudnCode-Linux-AnyDistro.run
First run
$ cd ~/engagements/acme-preview
$ audncode                 # alias: openclaude
  ✓ signed in as you@acme.com
  ✓ penclaw subscription: active
  ✓ identity check: verified
  ✓ marketplace tools: 14 loaded
  › model: pingu-unchained-10   (/model kong | godzilla)
  › /kairos on                  # autonomous loop
02 — What it does

An offensive-security engineer that lives in your terminal.

Everything PenClaw already gives you, plus a harness that writes and runs the code.

Auto Kairos mode

It keeps running.

Start an objective and walk away. The agent plans, recons, executes, reads the result, re-plans and retests on its own loop — it does not stop at the end of one tool call waiting for your next prompt.

PenClaw marketplace

Your installed tools just work.

Every marketplace item already installed on your PenClaw account is wired up automatically on first launch — MCP servers, scanners, custom tooling. Nothing to re-configure in the CLI.

Flat monthly

Unlimited tokens, no rate limiting.

One price per month instead of per-token metering. GPU inference is shared across the fleet, so the practical ceiling is your own machine — the CPU and RAM the agent runs its tooling on, not your token budget.

Terminal-native

A fork of Claude Code, built on OpenClaude.

The harness you already know — same file editing, shell, and MCP model — pointed at models that will do offensive-security work. Ships as audncode (alias openclaude) on Windows, macOS and Linux.

03 — Models

Switch engines mid-session.

Every model below is included in the flat monthly price. Change with /model without losing your session state.

Pingu Unchained 10

120B · Audn in-house

The uncensored model our own red team codes on. Long-context reasoning over exploit corpora, tuned on real pentester sessions. The default engine behind audncode.

Pingu Unchained KONG

derived from Kimi Code K2.7-abliterated

Code-first offensive engineering — exploit development, payload construction, tooling. Built for the part of the job where you are writing software, not prose.

Pingu Unchained GODZILLA

derived from Kimi Code K2.6 · Audn abliteration

Our proprietary abliteration method applied for offensive cyber. Heavier reasoning for chained attack paths where the first three ideas do not work.

Kimi K3 (abliterated)

newest addition

Complies with objectionable offensive-security requests under safeguards we consider more workable than the frontier providers, inside the KYC-gated perimeter.

Read the KONG & GODZILLA announcement
04 — Category

Red-team tools are functions. This is an agent.

The tools you already own get invoked — a scanner, a fuzzer, a payload generator behind a function call. They do their one job and hand control back. They never decide what to try next.

Execution model
Tools: A function the operator (or an orchestrator) calls. It runs once and returns.
audncode: An agent with its own loop: plan → recon → execute → read result → re-plan → retest, until the objective resolves.
Who decides the next step
Tools: You do. The tool has no view of the engagement beyond the arguments you passed it.
audncode: The agent does, from what it just learned about the target — and it carries that state across steps.
Coverage of the kill chain
Tools: One phase each. You stitch recon, exploitation and verification together yourself.
audncode: The whole chain in one session, including the retest that proves whether the fix actually held.
Refusals
Tools: Wrapped models refuse offensive tasks mid-engagement, or quietly water down the payload.
audncode: Uncensored models inside a KYC-gated perimeter — no refusal wall in the middle of authorized work.
Cost model
Tools: Per-seat, per-scan or per-token. Long autonomous runs get expensive fast.
audncode: Flat monthly, unlimited tokens, shared GPU. Long runs cost the same as short ones.
What limits you
Tools: Your licence tier and rate limits.
audncode: Your own hardware — the CPU and RAM on the box the agent is driving.

Practically: you are not buying capacity, you are buying an operator. The GPU is shared and the tokens are unlimited, so what bounds a run is the machine you point it at — its CPU, its RAM, and the scope you authorized.

05 — Access

Gated on purpose.

Uncensored models for offensive work only make sense behind a real identity gate. Three conditions, all enforced server-side on every request — not in the client you downloaded.

Active penclaw.ai subscription

Seats are limited. A free trial counts, and converts without reinstalling.

Government ID + selfie

A completed KYC identity check before the uncensored models unlock. No anonymous access.

Access is logged

Every request is attributed to a verified identity and retained. Use it on systems you are authorized to test.

Install it on the machine you actually work on.

Windows, macOS, Ubuntu/Linux. Unlimited tokens, no rate limiting, every Pingu Unchained model included — usable inside your free trial once your identity check clears.

Get audncode All Audn products