FAQ
Is AIGuard a SaaS product?
No. AIGuard is an open-source Python package you install into your own application and infrastructure. There is no cloud account to create.
Which LLM providers are supported?
AIGuard wraps LiteLLM, so any provider LiteLLM supports works — OpenAI, Anthropic, Google Gemini, Mistral, Ollama, vLLM, and more.
Does AIGuard call the LLM for me?
Only when you use aiguard.chat() or when use_live_model is enabled for an evaluation. Otherwise it operates on traces you already produced.
Is there a hosted dashboard?
No hosted version. The monitoring UI runs locally and reads from your own SQLite or Postgres database.
How are secrets handled?
API keys are read from environment variables named in aiguard.yaml via the api_key_env field. AIGuard never reads secrets directly from the YAML file.
Can I bring my own attack dataset?
Yes — point dataset_config at any JSON, JSONL, CSV, or HuggingFace dataset.
What is the overhead of the SDK?
The SDK is non-blocking. Calls are enqueued and processed by a daemon worker; latency overhead on the application thread is sub-millisecond.