What it does
Promptfoo is a CLI and library for evaluating and red-teaming apps built on language models. You describe tests in declarative config files, run them with promptfoo eval and look at the results with promptfoo view. It installs through npm, Homebrew or pip, or runs without installing through npx. It can compare models side by side, including OpenAI, Anthropic, Azure, Bedrock and Ollama, and can run checks automatically in CI/CD. Red teaming and vulnerability scanning probe an app for security problems and can produce vulnerability reports. A code scanning feature reviews pull requests for LLM-related security and compliance issues, and results can be shared with a team. The page says evals run entirely locally, so prompts do not leave your machine, and that it works with any LLM API or programming language. Most providers need an API key set as an environment variable. The page also states that Promptfoo is now part of OpenAI and remains open source under the MIT licence.
Capabilities
- Automated evaluations of prompts, agents and RAG apps
- Side-by-side comparison of models from many providers
- Red teaming and vulnerability scanning for LLM apps
- Declarative configs with CI/CD integration
- Code scanning of pull requests for LLM security and compliance issues
- Evals run locally
When to use it
- You want metric-based checks on prompt or model changes before shipping
- You need to probe an LLM app for security weaknesses
- You want LLM tests to run in your CI pipeline
- You want to compare several models on the same test cases
