KoppeK

Your offensive security battle station.

Automation, context, and AI — under your control.

KoppeK is a workflow layer for real engagements: run repeatable tasks, capture evidence automatically, and use AI to triage noisy output into actionable next steps — while you stay the operator.

One workspace for scope, runs, notes, and artifacts
Reproducible by default: every run saved, replayable, and exportable
AI summaries and hypotheses you can verify (not "one-click hacking")

Why KoppeK exists

Pentesting isn't hard because of tools — it's hard because of context switching. Terminals, scripts, notes, screenshots, payloads, chat logs… and then you still have to report.

Less chaos

Scope, runs, notes, and artifacts in one engagement workspace.

More rigor

Reproducible runs and an audit trail you can defend.

Faster decisions

AI turns noisy output into hypotheses and next steps you can verify.

New in Beta

BERSERK Mode

Multiple LLM backends run simultaneously — Claude, Codex, and Gemini each attack independently, then merge into a consensus plan with per-agent agreement scores.

  • Parallel multi-agent execution across all configured backends
  • Consensus-based plan merging with confidence scoring
  • Per-agent reasoning visible in the Agents tab grid
  • Pause / Resume — checkpoint any session mid-phase
BERSERK multi-agent mode in action

Multi-LLM Agent Architecture

Each configured backend receives the same target and objectives — they reason independently in parallel. KoppeK then merges their outputs into a single high-confidence attack plan. Where agents disagree, divergence scores surface uncertainty so you stay informed.

Agent Grid Watch each LLM's live reasoning side-by-side in the Agents tab
Consensus Merge Plans scored and merged — higher agreement means higher confidence in the next step
Backend Coverage Claude Code CLI, Codex CLI, OpenAI / Gemini / Claude API — all at once
BERSERK multi-agent mode showing parallel LLM execution
Advanced Feature

LLM-Driven Automation

When you're ready to scale, let the AI take the wheel — with proper authorization and full operator oversight.

Auto PWNR - LLM-driven automated penetration testing
🎯

Targeted Automation

Select which tools the AI can use and define scope boundaries before starting.

🧠

AI Reasoning Visible

Watch the LLM's thinking process in real-time — full transparency on every decision.

🛡️

Authorization Required

Exploitation requires explicit opt-in. Built for authorized testing only.

Supported LLM Providers

Claude Code Native integration
OpenAI API CodexCLI integration
Anthropic API Opus/Sonnet integration
Ollama Local models, check documentation
OpenRouter Any model, one API

Bring your own API keys. Your prompts, your control.

Auto-Generated Assessment Report

Every session produces a complete markdown report with findings, evidence, and AI reasoning — ready for your deliverables.

autopwnr-[REDACTED]-session.md Generated Report
# Auto PWNR Security Assessment Report

Target: [REDACTED].xxx.xxx.xxx
Duration: 6 minutes
Status: COMPLETE

## Executive Summary
| Severity | Count |
|----------|-------|
| Critical | 0     |
| High     | 0     |
| Medium   | 1     |
| Low      | 1     |

## AI Reasoning Log

🤖 Assistant:
"Initial scan reveals two open ports. Port 3389/tcp shows as 
'tcpwrapped' - TCP handshake completed but service closed 
connection. Could be host-based firewall or TCP wrappers.

I need more information. Let me run a more aggressive scan 
with additional scripts targeting these specific ports..."

✅ nmap - Targeted script scan on ports 25, 3389
Command: nmap -sV -sC --script=rdp-vuln-ms12-020 -p 25,3389

🤖 Assistant:
"Port 3389 now showing as 'filtered' rather than 'open'. 
This is inconsistent with the first scan — the host may have 
adaptive filtering. Let me try a wider port scan to find 
services on non-standard ports..."

✅ nmap - Wide SYN scan of top 10000 ports
✅ nmap - Full 65535 TCP port scan
✅ sqlmap - SQL injection on /api/users

## Findings

### 🔴 SQL Injection - Authentication Bypass
Severity: CRITICAL
Endpoint: POST /api/login
Payload: ' OR '1'='1' --

🤖 Exploitation:
"SQLi confirmed. Extracted 3 admin credentials from users table.
Recommending password reset and parameterized queries."
Alpha Complete

Alpha Testing Was a Success

We ran our initial Alpha with a select group of testers, and the results exceeded expectations. The feedback has been invaluable - we've gathered insights, squashed bugs, and refined the experience.

Stability improvements and faster container startup
Cleaner workflow: runs → notes → evidence in one place
Better exports: artifacts and history ready for reporting

What Alpha Testers Are Saying

Private Beta Countdown

🚀 Private Beta is coming.

We listened. We improved. Now we're ready for the next phase.

00 Days
:
00 Hours
:
00 Minutes
:
00 Seconds

March 27, 2026

Be among the first to experience the next evolution of offensive security tooling.

What KoppeK Actually Does

KoppeK isn't a list of tools. It's a workflow layer that keeps your assessment fast, organized, and reproducible.

Engagement Workspace Targets, scope, credentials, notes, and artifacts—organized like a real assessment.
Run Orchestration Launch tasks with consistent profiles, capture stdout/stderr, and keep history you can replay.
AI Triage & Summaries Turn noisy output into prioritized issues, hypotheses, and verification steps.
Evidence & Audit Trail Screenshots, logs, payloads, timestamps—ready for reporting and re-test.
Knowledge Layer Bring your own playbooks and internal notes; get suggestions grounded in your methodology.
Responsible Guardrails Operator confirmations, scope-aware behavior, and safety controls for authorized testing.
BERSERK Multi-Agent Claude, Codex, and Gemini attack in parallel — results merge into a consensus plan with per-agent scoring.
Playwright Automation AI-generated browser scripts for web app pentesting — describe a target, get a ready-to-run Playwright script.
Findings Graph & MITRE Entity graph connecting hosts, ports, CVEs, and MITRE ATT&CK techniques — auto-mapped from tool output.

You stay the operator. KoppeK keeps you fast, organized, and consistent.

From target to report—without the chaos

A clean loop you can repeat across engagements, with everything captured along the way.

1

Define Scope

Targets, rules of engagement, credentials, constraints

2

Run Discovery

Structured, repeatable profiles with live output

3

Triage

AI summary + your verification steps

4

Validate

Evidence capture + reproducible steps

5

Report & Re-test

Draft findings, remediation notes, replayable checks

Built for real engagements

Privacy-conscious by design: bring your own keys, keep artifacts in your environment, and stay in control.

Privacy & Control

  • Local-first workflow where possible
  • Bring-your-own API keys and models
  • Clear artifact storage paths
  • Exportable history and evidence

Built for Authorized Testing

  • Operator-in-the-loop confirmations
  • Scope reminders and guardrails
  • Designed for professional engagements
  • Not "one-click hacking"

FAQ

Is this "one-click hacking"?

No. KoppeK is operator-in-the-loop: it organizes workflows, captures evidence, and helps you triage faster — but you stay in control.

Do I need a powerful machine?

No. Tool execution runs in a container. AI can be remote or local depending on your setup and preferences.

Where does my data go?

Artifacts stay in your environment. If you enable AI, you bring your own keys and control what gets sent.

Get in Touch

Interested in KoppeK? Have questions? Let's connect.