Cloud Providers: OpenAI, Gemini, Anthropic¶

Configuration¶

Set provider API keys in your shell environment before running:

export OPENAI_API_KEY="sk-..."
export GEMINI_API_KEY="..."
export ANTHROPIC_API_KEY="sk-ant-..."
export GENAI_STUDIO_API_KEY="..."

Then reference models in your YAML config:

models:
  target_model: openai:gpt-4o
  attack_model: gemini:gemini-2.0-flash
  judge_model:  anthropic:claude-3-5-sonnet-20241022

Rate Limits and Retry¶

Cloud providers impose rate limits. The framework includes retry logic with exponential backoff in runner/models.py. Adjust max_retries and retry_backoff in config if needed:

runner:
  max_retries: 5
  retry_backoff: 2.0

Genai Studio (Google)¶

The genai: prefix targets Google's Generative AI Studio endpoint. The GENAI_STUDIO_API_KEY environment variable is required:

models:
  target_model: genai:gemma-3-27b-it
  attack_model: genai:llama3.3:70b

→ Back to Providers Overview