Multiple Red Teaming Modes

The dtx framework supports three flexible modes for running red team evaluations against language models. Each mode is designed for different levels of control—from beginner-friendly guided runs to fully customizable YAML-based test workflows.

Mode Comparison

Mode	Description	Best For
Guided Run	Interactive CLI wizard for setting up tests	Beginners, fast demos
Direct Run	Command-line based execution using flags (--agent, --dataset, etc.)	Developers, quick iterations
Advanced Run	Full YAML pipeline: scope → plan → execution	Teams, reproducible audits

---

Red Teaming Modes

Red Teaming Modes

├── 1. Guided Run
│   └── dtx redteam quick
│       - Interactive wizard
│       - Choose agent, dataset, evaluator
│
├── 2. Direct Run
│   └── dtx redteam run --agent <AGENT> --dataset <DATASET> [--eval <EVALUATOR>] [--url <MODEL>] [--keywords <KEYWORDS>]
│       ├── Example 1 (Airbench + IBM Eval):
│       │   dtx redteam run --agent echo --dataset airbench --eval ibm38
│       ├── Example 2 (Garak with built-in evaluator):
│       │   dtx redteam run --agent echo --dataset garak
│       ├── Example 3 (Keyword match):
│       │   dtx redteam run --agent echo --dataset beaver --eval any --keywords research
│       ├── Example 4 (HF model with evaluator):
│       │   dtx redteam run --agent hf_model --url arnir0/Tiny-LLM --dataset beaver --eval ibm38
│       ├── Example 5 (OpenAI model with Stringray):
│       │   dtx redteam run --agent openai --url gpt-4o --dataset stringray
│       └── Example 6 (Groq with LLaMA Guard model):
│           dtx redteam run --agent litellm --url groq/llama-3.1-8b-instant --dataset stringray
│
└── 3. Advanced Run (Scope → Plan → Run)
    ├── Step 1: Generate a scope file
    │       dtx redteam scope "test" test_scope.yml
    ├── Step 2: Generate a plan from scope
    │       dtx redteam plan test_scope.yml test_plan.yml --dataset stringray
    └── Step 3: Run the plan
            dtx redteam run --plan_file test_plan.yml --agent openai --url gpt-4o

---

🔧 Before You Run with Real Models

To run tests with providers like OpenAI, Groq, or Detoxio, make sure to create a .env file with your API credentials:

cp .env.template .env

Then open .env and fill in your keys, for example:

OPENAI_API_KEY=your-key
GROQ_API_KEY=your-key
HF_TOKEN=your-huggingface-token
LANGSMITH_API_KEY=your-key

🔑 Where to Get API Keys

Service	Purpose	Get API Key Link
OpenAI	Run models like gpt-4, gpt-4o	https://platform.openai.com/account/api-keys
Groq	Access fast LLaMA-3, Mistral models	https://console.groq.com/keys
Detoxio	Use Detoxio evaluators & policy LLMs	https://platform.detoxio.ai/api-keys
Hugging Face	Access gated models/datasets	https://huggingface.co/settings/tokens
LangChain Hub / LangSmith	Use prompt templates	https://smith.langchain.com/settings

---