Introducing self-service Giskard: open to beta testers #2868
henchaves
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi Giskard community👋
We've got some exciting news to share with you!
If you have run
vulnerability_scanfrom the open-source library and wanted to go further, now it's the time! The Giskard Hub, a product which carries our most complete red teaming scan, was built as an enterprise platform with onboarding designed for very large teams, making it harder for individuals and smaller teams to get started.We are changing that with a self-service version of the Hub, planned for the end of this year! 🚀
Before the launch, we are opening free access to a small group of beta testers, and we would like part of that group to come from the community.
What you do
You connect your agentic application to the Giskard Hub and run one red teaming audit. The Hub treats your agent as a black box and only needs an HTTP endpoint it can POST to, so the framework behind it (LangChain, LlamaIndex, plain Python, something vibe-coded) does not matter. It works with whatever request and response format your agent uses, whether you're building a chatbot, a classification system, a workflow automation agent, or something else entirely.
What you get back
You get a full red teaming audit report on where your agent fails. Each finding comes with the attack that worked, the conversation as evidence, a severity level, and guidance on how to fix it. The scan covers risks such as prompt injection, data privacy exfiltration, internal information exposure, excessive agency, harmful content generation, and hallucination.
What we ask in return
We ask for feedback on how the product feels when you use it on your own. We are especially interested in where the connection step got stuck, which findings you consider false positives, and what the report missed or failed to explain. There is no cost and no commitment beyond that.
How to join
Reply in this thread with three things:
If you would rather not describe your agent in public, send an email to blanca@giskard.ai instead. We will get back to you with access, and with a short call to talk through your use case first.
Thank you!
All reactions