Hawk-i Docs
Documentation/UNDER THE HOOD

AI Integration

LLM reasoning is optional and off by default. Enable it with the --ai flag. Hawk-i routes all model calls through litellm, so it works with several providers behind one interface.

Command-line flags

  • --ai - Enable LLM reasoning.
  • --ai-model - Specify the model, e.g. gemini/gemini-1.5-flash, openai/gpt-4, ollama/codellama.
  • --api-key - Provide the API key directly (alternative to environment variables).

The default model is gemini/gemini-1.5-flash.

Environment variables

Keys can be supplied via environment variables (recommended). Hawk-i routes to the right one by model prefix:

export GEMINI_API_KEY=your_key_here      # Google Gemini
export OPENAI_API_KEY=your_key_here      # OpenAI
export ANTHROPIC_API_KEY=your_key_here   # Anthropic

Your keys stay on your machine. Hawk-i passes them straight to the provider you chose and never routes them through any Hawk-i service.

Local models with Ollama

To keep everything offline, run a local model through Ollama. Specify the model as ollama/<model_name>:

$ hawki scan ./contracts --ai --ai-model ollama/codellama

Make sure Ollama is running locally and the model is pulled.

Fallback behavior

When --ai is not used, Hawk-i falls back to the explanation and impact templates provided by each rule. Every finding still ships with a clear description and remediation guidance, even with no LLM in the loop.

PreviousVulnerability LibraryNextPrivacy Promise