Skip to content

Model Providers

The hb providers commands manage the LLM providers Humanbound uses to generate adversarial attacks, run security judges, and produce evaluations. Six providers are supported out of the box (OpenAI, Anthropic Claude, Google Gemini, Azure OpenAI, Grok, and a Custom OpenAI-compatible option), plus an interactive setup flow. One provider can be marked as the default; experiments can override per-run with --provider-id.

List Providers

hb providers list

Add Provider (Interactive)

hb providers add -i

Add OpenAI Provider

hb providers add --name openai --api-key sk-...

Add Azure OpenAI Provider

hb providers add --name azureopenai \
  --api-key ... \
  --endpoint https://your-resource.openai.azure.com

Add Claude Provider

hb providers add --name claude --api-key sk-ant-...

Update Provider

# Update API key
hb providers update <id> --api-key sk-new-key

# Set as default provider
hb providers update <id> --default

Remove Provider

hb providers remove <id>

Supported Providers

Provider Label Requirements
Azure OpenAI azureopenai API key + endpoint URL
OpenAI openai API key
Anthropic Claude claude API key
Google Gemini gemini API key
Grok (xAI) grok API key
Custom custom API key + whitelisted endpoint

Custom providers

For self-hosted or OpenAI-compatible models, the endpoint must be whitelisted via the CUSTOM_MODEL_PROVIDER_ENDPOINT environment variable in your Humanbound deployment.