Promptfoo

Promptfoo is an open-source testing and evaluation framework designed to help developers systematically test LLM applications, prompts, and AI agent behaviors. It provides a CLI-driven workflow for defining test cases, running evaluations across multiple models and prompt variants, and comparing results with automated scoring — essential for building reliable AI agents that behave predictably in production.

The framework supports a wide range of assertion types including exact matching, semantic similarity, model-graded evaluations, and custom JavaScript/Python assertions. Developers can test across multiple LLM providers simultaneously, comparing how different models handle the same prompts and scenarios. This is particularly valuable for agent development where choosing the right model for each task is critical.

Promptfoo's automated red-teaming capability is a standout feature for agent security. It can automatically generate adversarial inputs to test agent robustness against prompt injection, jailbreaking, data exfiltration, and other attack vectors. This helps developers identify and fix agent vulnerabilities before deployment.

The framework integrates with CI/CD pipelines, enabling automated testing of agent behaviors on every code change. Results are displayed in an interactive web UI that makes it easy to compare outputs, identify regressions, and track improvements over time. Promptfoo supports all major LLM providers including OpenAI, Anthropic, Google, AWS Bedrock, and local models via Ollama. With its focus on practical testing workflows, Promptfoo has become the most popular open-source tool for LLM evaluation.