dtx 0.17.0

AI Red Teaming Tool and Framework

How to Install dtx

Before you begin, choose how you want to run dtx depending on your environment and requirements.

---

Option 1: Install dtx locally with full dependencies (torch etc.)

This is recommended if you plan to run local models (e.g., Hugging Face, Ollama) on your machine.

pip install dtx[torch]

Includes:

• Core CLI
• torch, transformers for local LLM and classifier execution
• Supports all datasets and local execution

---

Option 2: Install dtx if torch is already installed

If your environment already has torch installed (for example, in a GPU-accelerated ML environment), you can skip extras:

pip install dtx

dtx will use your existing torch installation.

Tip: Verify torch is installed:

python -c "import torch; print(torch.__version__)"

---

Option 3: Use uv for fast installation in a clean environment

If you're creating a new environment and want fast dependency resolution with uv:

Install uv

curl -LsSf https://astral.sh/uv/install.sh | sh

Install dtx with full dependencies

uv pip install dtx[torch]

---

Option 4: Use Docker wrapper (ddtx)

If you prefer Dockerized execution (no local torch install required), you can use ddtx.

1. Install dtx (for the ddtx wrapper CLI):

pip install dtx

2. Use ddtx to run inside Docker:

ddtx redteam scope "Describe your agent" output.yml

Features:

• No need to install torch locally
• Fully containerized execution
• Automatically mounts .env and working directories
• Use Docker-managed templates and tools

---

Summary of Options

Method	Use case	Install command
Local, full dependencies	Full feature set, local models	pip install dtx[torch]
Local, existing torch	You already have torch installed	pip install dtx
New env, fast install	Clean, fast setup	uv pip install dtx[torch]
Docker (ddtx)	No local Python dependencies, isolated	pip install dtx + use ddtx CLI