HomeSolutions
Find out what your assistant can be talked into.
Nullgap attacks your chatbot or agent with 32 prompt-injection attacks and proves it when the app leaks a secret, reveals its instructions or calls a tool it must not.
Attacks a real attacker would try, with proof when one works.
12 categories of attack
Instruction override, fake system messages, system prompt extraction, secret disclosure, indirect injection, tool misuse and more.
A planted canary secret
Nullgap plants a canary secret, so a leak is proven by the secret coming back, not by a guess.
Your API, as it is
Point it at a chat-API-compatible endpoint or describe a custom JSON API.
Gentle and bounded
At most 4 requests a second and a request budget of up to 500. The most it can use is shown before it starts.
Header secrets sealed
API keys you add as headers are sealed and erased after the run.
Only apps you may test
Each run needs you to confirm: “I own this app or am authorised”.
The 12 categories.
All mapped to the OWASP Top 10 for LLM apps.
Where to start
AI app questions
What kinds of AI apps can be tested?
Any app reachable through a chat-API-compatible endpoint or a custom JSON API, such as a support chatbot or an agent with tools.
Will it overload our app?
No. It sends at most 4 requests a second, within a request budget of up to 500 that you set.
How is it priced?
AI app tests use credits, and the most a run can use is shown before it starts. AI app testing is included on Enterprise.
What happens to the API key we give it?
Header secrets are sealed for the run and erased after it.
Get your first proof today.
Connect a repository and watch Nullgap find, prove and fix. You only ever see what it could prove.
Free 14-day trial · No card · or book a demo with a security engineer