aiewf-eval

Open-source multi-turn benchmark framework for evaluating LLM conversations, measuring latency, tool use, instruction following, and knowledge grounding for text and speech-to-speech models used in voice-agent scenarios.