Hire an autonomous AI pentester that works on its own to keep your targets secure, 24/7, no sleep, minimal supervision.
Point it at a target and it handles the rest — recon, scanning, exploitation, and reporting. No content filters. No refusals. Runs on dedicated GPU infrastructure. Connect via WhatsApp, Telegram, Discord, or API.
$20/mo for Pro, $999/mo for Ultimate includes 1 week unlimited tokens (see audn.ai/necromicon) — $100 of extra AI usage a month included for both packages
An AI agent that lives in your workspace — scanning, testing, and reporting vulnerabilities around the clock.
Built on Pingu Unchained 10 Abliterated LLM. No safety theater. No refusals. Real offensive security analysis without artificial limits.
Deploy any cybersecurity MCP server from GitHub in one click — Nmap, Nuclei, SQLmap, Burp Suite, and 50+ more. Paste a repo URL at pingu.audn.ai/mcp, deploy to Vercel or Render, and it's live. No rejections. Always works.
Subdomain enumeration, port scanning, service fingerprinting, and vulnerability detection — all driven by conversational commands.
Run pentests from Signal, Slack, Discord, Telegram, or WhatsApp. Just message your agent — it handles the rest.
Full Cursor-like coding experience in the cloud on your own tenant. Edit configs, write exploits, and manage your instance from a browser-based IDE with complete filesystem access.
An AI agent with sudo access that retrofits your OpenClaw instance in real time. Install tools, patch configs, deploy MCP servers, and reconfigure your entire stack — conversationally, from the console.
Running on NVIDIA H100 with 80GB VRAM. 262K context window. Flash attention. Fast inference at ~195 tokens/sec. No token limits — use as much as you need.
AI coding agent on the Claude Code harness — powered by Pingu Unchained 10 with $100 of usage a month, included in your $20/mo Pro plan ($999/mo for Ultimate). Bring your own API key to swap LLMs anytime. Kairos Mode runs 24/7 on your always-on cloud instance. Personalize your agent easily with Meta-Claw.
Get structured vulnerability reports with severity ratings, proof-of-concept payloads, and remediation guidance — delivered right in chat.
OpenAI-compatible API endpoint included. Integrate pentesting into your CI/CD pipeline or build custom security workflows.
OpenClaude, packaged as a native one-click installer. Ships the audncode command (alias openclaude) on the Claude Code harness, powered by Pingu Unchained 10 — no safety theater, no refusals.
Requires an active penclaw.ai subscription and a completed government KYC identity check, enforced server-side on every launch. First run signs you in.
Your PenClaw instance runs 24/7 in the cloud — making it the perfect home for Claude Code's Kairos Mode. An always-on AI coding agent that never sleeps, powered by Pingu Unchained 4 with $100 of usage a month included.
$100 of usage a month is included in your $20/mo Pro plan (or $999/mo Ultimate); the free trial includes $20. Pingu Unchained 10 runs on the Claude Code harness — top up in $20 packs when you need more.
Claude Code Kairos Mode works perfectly here because your instance runs 24/7 in the cloud. Schedule autonomous tasks, let the agent work overnight, and wake up to results.
Want to use a different LLM? Bring your own API key to swap models anytime. Pingu Unchained 10 stays available as your default.
Use Meta-Claw to customize your OpenClaude instance — install tools, configure settings, and tailor the agent to your workflow, all conversationally.
Don't want the cloud instance? AudnCode installs natively and works the same on
your own PC, VM, or Mac — same audncode command, same Pingu
Unchained 4.
Get a full Linux desktop environment streamed directly to your browser. Run GUI security tools, web browsers, file managers, and anything else you need — all in the cloud.
Full Linux Desktop
No GPU Required
Run GUI Tools
Browser Included
OpenClaw shows you what it's doing — it operates the cloud desktop alongside you in real time. When it needs your help resolving CAPTCHAs or confirming actions, it notifies you directly.
All included — OpenClaude with Pingu Unchained 10 ($100 of AI usage a month), Claude Code Kairos Mode, terminal, file editor, desktop GUI, and all security tools. BYOK to swap LLMs.
Pre-loaded cybersecurity MCP servers ready to install. Or deploy any MCP repo from GitHub in one click — no rejections, always works.
Nmap
Reconnaissance
Nuclei
Offensive
SQLmap
Offensive
Burp Suite
Offensive
Semgrep
Defensive
Shodan
Reconnaissance
Amass
Reconnaissance
BloodHound
Offensive
Ghidra
Reverse Eng.
ScoutSuite
Cloud Security
VirusTotal
Threat Intel
MobSF
Mobile
Found an MCP server on GitHub? Paste the repo URL at pingu.audn.ai/mcp and deploy it to Vercel or Render in one click. No config, no Docker, no rejections — it just works.
Cloud Desktop GUI
Full Linux desktop streamed to your browser. Run GUI tools, browsers, and file managers. OpenClaw operates it alongside you.
Cursor IDE in the Cloud
Full coding environment on your own tenant. Edit files, write scripts, manage configs from a browser-based IDE.
Web Console
Full terminal access to your instance. Run tools, inspect logs, execute commands — all from the browser.
Meta-Claw Agent
An AI agent with sudo that retrofits your OpenClaw instance. Install packages, deploy MCPs, patch configs — conversationally.
OpenClaude
AI coding agent on the Claude Code harness with Pingu Unchained 4 — $100 of usage a month included. Kairos Mode runs 24/7 on your cloud instance. BYOK to swap LLMs. Personalize with Meta-Claw.
Skills Marketplace
50+ pre-loaded security MCPs. One-click install for hosted servers. One-click deploy for self-host repos via Audn.AI.
Find MCP on GitHub
Any MCP server repo
Deploy to Vercel/Render
One click, no config
Paste URL in PenClaw
Instant connection
PenClaw ships as a pentesting preset — but your instance
is a full cloud AI workspace. Use reset-identity to wipe the
pentester persona and build something completely different. Automate workflows, run bots, build agents — it's
your machine.
Here's what people have already built:
Full-scope penetration testing of a YC S25 startup — web application, authentication flows, backend LLM, voice AI, and behavioral testing. All in one day.
Done via OpenClaude on /console orchestrating the OpenClaw instance for multi-subagent kickoff. OpenClaude orchestration keeps getting better — we're launching OpenClaw as an installable MCP inside cloud OpenClaude on /console with Kairos Mode enabled.
Using Computer Use on the cloud desktop, we ran a LinkedIn job applier bot that browses, fills forms, and submits applications autonomously. Point it at job listings and let it work.
A Grok-like AI responder that monitors mentions, answers cybersecurity questions, manages threads, and tweets about the latest OpenClaw vulnerabilities. Has per-user personal memory — remembers who you are across conversations.
Reddit didn't exactly allow this — we built it anyway. Posts, comments, engages in threads, and works exactly like the X bot. Same personal memory system, same autonomous operation.
You can obviously do so many things with this. An always-on cloud instance with $100 of AI usage a month, computer use, a full desktop, and a Claude Code harness — the possibilities are endless.
Email us or tap the blue chat button on the bottom left and tell us what you're building. We'll list your use case right here.
Two plans. Full access. $100 of AI usage a month. No hidden fees.
Single Dedicated GPU
Assigned exclusively to you
Up to 50 Users in Parallel
Shared GPU · hover for details
$100 of usage a month
$0.54/M in · $3.84/M out · $20 top-ups
Everything. OpenClaude included. $100 of AI usage a month.
$20/mo. Cancel anytime.
Identity verification required. PenClaw runs a highly potent, uncensored security model, so — like OpenAI and Anthropic do for their most capable models — we grant product access only after a one-time selfie + government ID check. This applies even during the free trial and protects our 50+ verified customers. Prefer not to verify? You can cancel anytime from the billing portal.
$20/mo — $100 of AI usage a month included.
Maximum power. $1,000 of Kimi K3 (refusals removed) credits every month.
Limited early access — only 50 seats
$999/mo after trial. Cancel anytime.
Identity verification required. Like every PenClaw plan, Ultimate grants access to a highly potent, uncensored model only after a one-time selfie + government ID check.
Every PenClaw plan includes two uncensored models with $100 token limits. Two higher-intelligence frontier models are available as monthly add-ons, with usage limits.
Listed from highest to lowest intelligence. Switch models any time with
/model.
Our most capable cyber model — a Kimi K3 based model with audn abliteration. Built for the hardest offensive-security reasoning, long multi-step engagements and large codebases. Live on the audncode CLI and penclaw.ai.
A Kimi K2.6 based model with audn abliteration. Deep reasoning over long files and repos, with a large context window.
A QWEN3.8 abliterated model. Fast, high concurrency, uncensored — a strong everyday driver, drawn from your $100/mo allowance.
Built on Qwen3.8‑abliterated — the base model included in every plan, and the default on the OpenClaude harness. $100 of usage a month included ($20 on the free trial); top up later is available. 1 identity can only have 1 penclaw account in it's lifetime. Only one trial in lifetime is available.
Add-ons stack on your Pro plan. Necromicon and GODZILLA are separate monthly subscriptions on top of an active Pro plan, each cancellable on its own. Buying one does not require the other. Ultimate includes both — no add-on needed; its Kimi K3 usage draws on $1,000 of credits per billing month.
Identity verification required. Every model here is highly potent and uncensored, so access is granted only after a one-time selfie + government ID check. Need them without KYC? The same models are available pay‑as‑you‑go at platform.audn.ai, and audn.ai/audncode works without verification too — no KYC needed on either.
The exact ceiling on every tier. Windows reset on their own — and no plan has a daily or hourly token limit.
| Limitwindow · reset | Baseno add-on | Godzilla‑99+$79/mo | Necromicon‑179+$179/mo | Ultimatefast week · $999/mo | Ultimatestandard week |
|---|---|---|---|---|---|
| Included usage on pingu-unchained-10 & KONGcalendar month · resets on the 1st, 00:00 UTC | $100 / month$20 on the free trial · top up in $20 packs | $100 / month | $100 / month | $100 / month | $100 / month |
| Requests / hourresets at the top of the hour | 6,000 | 7,200 | 26,500 | No cap | No cap |
| Kimi K3-class usageNECROMICON / WARLOCK · resets Monday 00:00 UTC | Credit-metered$0.54/M in · $3.84/M out | 23,000,000 | 15,250,000 | No capunlimited tokens | $1,000/mo credit$4.50/M in · $0.40/M cached · $4.50/M out |
| Input tokens / monthpaid-tier serving only · resets on the 1st | $0.54 per 1Mdrawn from the allowance | 40,000,000 | 30,000,000 | No cap | $4.50 per 1M$0.40/M cached · from the $1,000 credit |
| Output tokens / monthresets on the 1st | $3.84 per 1Mdrawn from the allowance | 6,000,000 | 500,000 | No cap | $4.50 per 1Mfrom the $1,000 credit |
| Any daily or hourly token limit | None | None | None | None | None |
| Data retention and trainingopt-in by default | Opt-in (default) | Opt-in (default) | Opt-in (default) | Opt-out available | Opt-in (default) |
Add-on columns cover usage of that add-on’s model. Ultimate includes every model with no per-window cap; instead its K3-class usage is $1,000 of credits per billing month, metered at $4.50/M in · $0.40/M cached · $4.50/M out. Requests served in the cohort fast week are not counted. The $20 plan includes $100 of usage a month on pingu-unchained-10 and KONG ($0.54 / M input, $3.84 / M output); the trial includes $20. Top up in $20 packs from the dashboard. Add-on models continue past their allowance on top-up credit at their own card (Bartzabel $0.54 / $3.84; NECROMICON, QWEN38 and WARLOCK $4.50 / $0.40 cached / $4.50).
On standard-speed weeks, Kimi K3 runs as one of three two-leg pipelines; Omen Offensive Cyber is our stealth model. Delivered = real harmful content returned; Refused = substantive refusals — LLM-judge, full 519 non-streaming (benchmark).
| Variantmodel id | Alias / served by | Pipeline | Deliveredreal content, ns | Refusedsubstantive | Serving |
|---|---|---|---|---|---|
| k3-2leg-venicek3-thinker-qwen38 and similar | Qwen38 answerer | K3 thinker → qwen38 answerer | 66.6% | 18.8% | 1M input262K output max · QWEN38 answer from Audn’s own H100, Venice.AI and Wiro |
| k3-2leg-double-ven | WARLOCK | K3 thinker → qwen answer folded in as reasoning → K3 final answer | 71.9% | 0.0% | 1M input · QWEN38 side answered by Audn’s own H100, Venice.AI and Wiro.ai |
| k3-2legK3 → K3 | NECROMICON | Both thinker and answerer legs are Kimi K3 | 61.7% | 1.5% | 1M input · NECROMICON answer from our modal.com shared endpoint |
| Omen Offensive CyberStealth | OMEN | Offensive-cyber stealth model | 50.7% | 17.4% | — |
Standard-speed weeks only. The K3 leg is hosted on modal.com; the qwen answerer leg runs on Audn’s own H100, Venice.AI and Wiro.ai to keep availability high. During a fast week Ultimate serves unlimited K3 with no two-leg metering (see the plan-limits table above). Some people used $36K worth of tokens with $999 subscription last cohort.
Ultimate’s fast week serves KIMI‑K3‑MXFP4‑DERISKED‑V2 single-stage on 8×B300 — one request thinks and answers. Standard weeks run the same weights as a two-leg Modal pipeline. Same phrasing score, but the single stage is ~12 points more compliant on real content (LLM-judge benchmark).
| MetricLLM-judge, 520 harmful prompts | Fast weekKIMI-K3-1M · single stage | Standard weekModal K3REL · two-leg |
|---|---|---|
| How it runs | One request thinks & answers | K3 thinker → answerer (two legs) |
| Hardware | 8×B300 | modal.com |
| Throughput | up to 666 tok/sec | standard speed |
| Regex complyphrasing classifier | 97.1%505/520 | 96.7%503/520 |
| Deliveredreal harmful content | 76.7%399 | 65.0%338 |
| Deflectedsafe substitute / reframe | 17.5%91 | 19.6%102 |
| Refusedsubstantive | 5.8%30 | 15.0%78 |
Both servings read ~97% on the phrasing classifier, so it cannot tell them apart; the LLM-judge (substance, not tone) shows the fast single-stage endpoint delivers real content 77% vs 65% and truly refuses 6% vs 15% — a ~12-point gap. Warlock (GLM 5.3) benches at 82.5% delivered / 7.5% refused. Source: audn-ai/refusal-benchmark.
Every penclaw.ai plan — cloud instance, API, and audncode alike — requires a one-time identity check: government ID plus a selfie, via Persona, including during the free trial. If that’s not for you, platform.audn.ai serves the same models pay‑as‑you‑go and requires no identity verification.
API and audncode only. Cloud instances stay on penclaw.ai, and stay verified.
1. Install Featured MCP Servers — Start with the four featured items in the Marketplace: Audn Sec QA (free, voice & text agent red-teaming), HOPX Sandbox (HOPX managed, cloud code execution), Supermemory (free, persistent memory for the agent), and MCP Server Manager (free, manage all your MCP connections). These are one-click installs and give you the core toolkit.
2. Understand What's Free — All skills in the marketplace are free. Hosted MCP servers (like Audn Sec QA and Supermemory) are free. HOPX Sandbox is managed by HOPX. For self-host MCP servers, you get a 1-day free trial with one-click deployment to Render.
3. Use Supermemory for Persistence — Install Supermemory so your agent remembers previous scans, targets, and findings across sessions. This builds a knowledge base that makes each subsequent pentest smarter.
4. Run Continuous Scans — Point the agent at a target and let it run autonomously. Use the Console to monitor progress, and check Session Logs to review findings. The agent works best when given clear scope and left to enumerate, scan, and report.
5. Pair Recon with Action — After enumerating emails or subdomains, ask the agent to cross-reference with OSINT sources, check for leaked credentials, or test authentication. Chain multiple tools together for deeper coverage.
6. Deploy More MCP Servers — Browse the marketplace for 50+ security MCPs. Deploy any GitHub MCP repo in one click via pingu.audn.ai/mcp.
Deploy an abliterated AI agent with $100 of AI usage a month into your workplace in under 60 seconds. Operate it from any channel.
$20/mo for Pro, $999/mo for Ultimate includes 1 week unlimited tokens (see audn.ai/necromicon) — $100 of extra AI usage a month included for both packages