🫡 Introducing Nikki: Your Offline, Uncensored AI Red Team Assistant in the Terminal

Source: Dev.to

No cloud. No censorship. No hallucinated APIs.
Just battle‑tested code, curated from real offensive repos, served right in your shell.

As offensive security professionals we’ve all been there:

• You need a working AMSI bypass now.
• You’re knee‑deep in a lab or authorized engagement.
• You can’t afford to wait for slow web UIs… or worse — leak prompts to the cloud.

That’s why I built Nikki — a local, CLI‑first AI teammate that:

• Runs 100 % offline via Ollama.
• Uses uncensored base models (rnj-1:latest).
• Answers questions using real knowledge from PayloadsAllTheThings, Atomic Red Team, SharpCollection, and more.
• Speaks your language.

…and never says “I can’t assist with that.”

Why Nikki Is Different

Most “AI pentest assistants” are:

• Hosted in the cloud → prompt leakage risk.
• Fine‑tuned with RLHF → ethical filters block offensive techniques.
• Trained on generic data → hallucinate fake APIs (e.g., WshShell.ConnectObject()).

Nikki avoids all three.

It’s built on:

• aichat – Rust, CLI‑native, sessions, RAG.
• Ollama – local LLM + embedding via nomic-embed-text.
• Real offensive repos – cloned, cleaned, indexed.
• A strict system prompt that forbids guessing.

“If it’s not in the source repos, I don’t know it.” — Nikki (probably)

Architecture: AI That Knows Where It Gets Its Info

┌──────────────┐
│   You:       │
│   Nikki how  │
│   to bypass  │
│   AMSI?      │
└──────┬───────┘
       ▼
┌──────────────┐     ┌──────────────────┐
│   aichat     │────▶│   RAG over       │
│   (CLI)      │     │   ~/rag-data/    │
└──────┬───────┘     │   • Payloads...  │
       │             │   • SharpColl... │
       ▼             │   • Atomic Red   │
┌──────────────┐     └─────────┬────────┘
│   Ollama     │◀──────────────┘
│   • LLM:     │
│     deepseek-│
│     coder    │
│   • Embed:   │
│     nomic-   │
│     embed    │
└──────────────┘

Every answer is grounded in real tools, and you can always verify the source:

> .sources rag

Quick Start (Athena OS / Arch)

# 1. Install dependencies
sudo pacman -S aichat ollama git fish

# 2. Enable Ollama
systemctl --user enable --now ollama

# 3. Pull models
ollama pull deepseek-coder:6.7b-base
ollama pull nomic-embed-text

# 4. Install Nikki
git clone https://github.com/toxy4ny/nikki-ai.git
cd nikki-ai && makepkg -si

# 5. Load knowledge
setup-rag

# 6. Ask anything
Nikki generate a C2 beacon with XOR encryption

Real Usage Examples

One‑off query

Nikki show me AMSI bypass from PayloadsAllTheThings

Multi‑turn session

Nikki --session c2 "Write a reverse TCP shell in C"
Nikki --session c2 "Add process hollowing"
Nikki --session c2 "Compile with mingw for x64"

Verify sources

aichat
> .rag nikki-kb
> How does Unicorn do DDE attacks?
> .sources rag   # ← shows exact file from trustedsec/unicorn

What’s Included

Ethical Note

Nikki is designed only for:

• Authorized penetration tests.
• Red‑team exercises.
• Closed‑lab education (HTB, PWN, etc.).

It does not generate novel exploits — only techniques already public in trusted repositories.
Always review code before execution.

Why This Matters

In an era where:

• Cloud AI logs every prompt.
• “Ethical AI” blocks real red‑team techniques.
• Hallucinations waste precious engagement time.

Nikki restores control to the operator.
She’s not a chatbot; she’s your offline, open‑source, truth‑grounded AI teammate.

Try It Today

Made with ❤️ for the offensive security community.

“The best AI for red teaming is the one that never phones home.”
— toxy4ny, 2025