Promptfoo is a command-line tool and Node.js library for evaluating and red-teaming LLM applications. Evaluations run a set of prompts and test cases against one or more models and score the outputs, so prompt and model changes are compared on metrics rather than by trial and error. Results appear as a side-by-side matrix in a local web viewer or on the command line, and tests are configured in YAML.
Red teaming generates attacks tailored to the target application (direct and indirect prompt injection, jailbreaks, data and PII leaks, business-rule violations, insecure tool use in agents, and toxic content) and produces a vulnerability report. Code scanning reviews pull requests for LLM-related security and compliance issues.
Evals run locally, so prompts stay on the machine, and the tool works with any LLM API or programming language. It compares models from OpenAI, Anthropic, Azure, Bedrock, Ollama, and other providers, caches results, reloads on change, and runs as a check in CI/CD pipelines.
Features
- Automated evaluations: test prompts and models against defined test cases and assertions
- Model comparison: side-by-side output matrices across providers such as OpenAI, Anthropic, Azure, Bedrock, and Ollama
- Red teaming: generated attacks for prompt injection, jailbreaks, PII leaks, business-rule violations, and insecure agent tool use
- Vulnerability reports: summarizes red-team findings for an application
- Code scanning: reviews pull requests for LLM-related security and compliance issues
- CI/CD checks: runs evals and scans automatically in a pipeline
- Local execution: evals run on the developer's machine with caching and live reload
- Web viewer: browse and share evaluation results with a team
