promptmap
A free tool that automatically tries to trick your own AI chatbot into misbehaving — leaking its instructions, saying something harmful, or ignoring its rules — so you find those weaknesses before a real user does.
🔗 Visit promptmap
Description
An AI application built on a system prompt (the hidden instructions telling it how to behave) can often be manipulated into ignoring those instructions through carefully worded "prompt injection" attacks. promptmap automates the process of testing for this: it sends a library of attack prompts to a target LLM and uses a separate "controller" LLM to judge whether each attack actually succeeded, rather than requiring a human to review every response by hand.
It supports two testing modes — white-box, where you give it your system prompt and let it run the target model directly, and black-box, where you point it at a live HTTP endpoint and it attacks from the outside, the way an actual attacker would approach a deployed app. It ships with 50+ pre-built test rules across six categories (prompt stealing, jailbreaking, harmful content, bias, and more), supports OpenAI, Anthropic, Google, XAI, and Ollama models, and lets you write custom rules in YAML.
💬 Our review
The short version: if you're shipping an LLM-powered feature with a system prompt worth protecting, promptmap is a free, quick way to stress-test it before launch instead of finding out about a jailbreak from a user's screenshot on social media.
It covers similar ground to commercial AI red-teaming platforms like General Analysis or Stingrai, but as a free, open-source CLI tool rather than a managed service — you get the same core idea (automated adversarial prompts, pass/fail evaluation) without the enterprise dashboard, human validation, or ongoing monitoring those platforms add. For an indie developer or small team, promptmap is the practical starting point; for an enterprise needing continuous monitoring and human-verified findings, it's worth pairing with (or upgrading to) a dedicated commercial platform.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Licence GPL-3.0, gratuit, self-hosted.
Pros
Gratuit et open source, 50+ règles de test prêtes à l'emploi
Mode white-box ET black-box (test d'un endpoint HTTP réel)
Compatible OpenAI, Anthropic, Google, XAI, Ollama
Cons
Pas de tableau de bord ni de monitoring continu comme les plateformes commerciales
Nécessite de lancer et interpréter les tests soi-même
Pas de validation humaine des résultats (contrairement à Stingrai)