Promptfoo
Automated model evaluations with ready made attack strategies and reports.
About Promptfoo
Promptfoo states that it helps developers and enterprises build secure, reliable AI applications. Promptfoo is now part of OpenAI, based in San Francisco, California. Promptfoo remains open source and continues to build tools for secure, reliable AI applications.
Its red teaming product tests for 50+ vulnerability types that target a specific application, from prompt injections to data exfiltration. Findings can be mapped to OWASP, NIST, EU AI Act, and custom frameworks for audit-ready reporting. The red teaming product can be deployed in the cloud or on-premises.
Drafted from promptfoo.dev · not yet confirmed by Promptfoo
How it works
Promptfoo is a test runner for LLM apps that also does red teaming. You write a config pointing at your model or endpoint, choose plugins (harmful content, PII leakage, prompt injection, jailbreaks, bias, hallucination, tool misuse) and strategies (multi turn, encoding, role play), and it generates attacks, runs them, grades results and produces a report you can gate a release on. It runs locally, in CI, and has a web viewer.
It's the most popular open tool for this because it's fast to set up and fits developer workflows. There's a hosted enterprise version with team features.
From the project
From the project
Reviews
Be the first to review it
AnonymousKnow a team weighing Promptfoo? Send them this page.