Autonomous pentesting with Docker sub-agents and a live terminal dashboard.

Kratos is a standalone CLI rewrite inspired by the concept of hacker-ai. It is not a code fork of that repository — only the core idea (coordinator + Docker sub-agents for autonomous pentesting) carries over.

• Python 3.13+
• uv
• Docker Desktop / Docker daemon running
• An LLM API key (Groq, OpenAI, Anthropic, or Google AI Studio)

cd cli
cp .env.example .env
# Edit .env and set an API key for your chosen provider
uv sync

# Run a scan (opens live dashboard at http://127.0.0.1:8765)
uv run kratos scan 192.168.1.1

# Multiple targets, custom vectors, JSON output
uv run kratos scan 10.0.0.5,https://example.com \
  --vectors "port scan,web scan" \
  --note "lab environment only" \
  --output results/findings.json

# Headless (no web UI)
uv run kratos scan 192.168.1.1 --no-ui -o findings.json

# Custom dashboard port
uv run kratos scan example.com --port 9000 --no-browser

# Pick LLM provider explicitly
uv run kratos scan 192.168.1.1 --provider openai
uv run kratos scan 192.168.1.1 --provider anthropic
uv run kratos scan 192.168.1.1 --provider openrouter

Kratos auto-detects the provider from your environment. Set one API key, or set LLM_PROVIDER / --provider when multiple keys are present. Auto-detect priority: openai → anthropic → google → openrouter → groq.

Override models with LLM_MAIN_MODEL and LLM_SUB_MODEL.

When a scan starts, a local FastAPI server serves a dashboard with:

• Per sub-agent status, completed steps, and findings
• Live xterm.js terminal output from each Docker sandbox
• WebSocket updates (~500ms refresh)

Press Ctrl+C in the terminal to stop the scan and clean up containers.

Only scan systems you own or have explicit written permission to test.

Kratos is open source. You are free to use, modify, and contribute to this project.

Contributions are welcome — open an issue or pull request if you want to improve Kratos.

Licensed under the MIT License.