Learn
LoginRegister
← All tools

Garak

Tool manual · Kali & common usage

AI Security · Intermediate · AI Security lane

IntermediateAI Security
Authorized usage only. Capture and inspect traffic only on systems and networks you own or have written permission to analyze.

What is Garak?

Garak probes LLM applications for weaknesses (prompt injection, leakage, …).

AI red-team style testing of models/apps you own.

Official website: Garak · Documentation

When do you use it?

  • You need LLM vulnerability probing framework.
  • You are following a CYNARKS lab or authorized engagement that calls for Garak.
  • You already tried simpler checks and need this tool’s deeper capability.

Who uses it?

  • Security practitioners
  • Students in authorized labs
  • Engineers validating their own systems

How do you use it? (typical flow)

  1. Install Garak using the steps below (prefer Kali package or official installer).
  2. Confirm you have authorization for the target (lab VM, own system, or written scope).
  3. Run --help / man (or the Help section) to see current flags for your version.
  4. Start with a safe example from Basic commands against localhost or an authorized lab target.
  5. Record what you ran, what you observed, and any follow-up (filter, report, remediation).

Help & manual (like man / --help)

Every serious tool has a built-in way to look up flags. Learn these first so you are not stuck memorizing every option.

  1. 1. Help

    garak --help

  2. 2. Repo

    https://github.com/leondz/garak

Kali Linux

pip install garak — modern AI security toolkit usable from Kali.

1. Installation

OSInstallation
Kali Linuxpip install garak — modern AI security toolkit usable from Kali.
pippip install garak

2. Basic commands

  1. 1. Test your own LLM app endpoint

    garak --model_type openai --model_name gpt-4o-mini

    Test your own LLM app endpoint

  2. 2. Show help

    garak --help

    List flags for your installed version (same habit as man/--help).

  3. 3. Short help

    garak -h

    Alternate help flag when supported.

5. Useful features

  • Probes
  • Reports

6. Practical examples

  1. Example 1: Test your own LLM app endpoint

    garak --model_type openai --model_name gpt-4o-mini

    Test your own LLM app endpoint

  2. Example 2: Show help

    garak --help

    List flags for your installed version (same habit as man/--help).

  3. Example 3: Short help

    garak -h

    Alternate help flag when supported.

7. Important notes

  • Red-team LLMs you control.
  • Authorized targets only — labs, systems you own, or written permission.
  • Prefer reading --help / man / official docs for your installed version; flags change over time.

Related course: ai security · web application security · Commands: /commands