For security engineering
One base URL, and every call is governed.
Point your SDK at an OpenAI-compatible endpoint, or call the scanning API directly. Policies are plain English, agents run inside a blast radius you set, and campaigns are scheduled.
Proxy · REST API · blast radius
Exposure
What you have to wire up.
- W-01
Enforcement in the path
You need inline scoring on real traffic without rebuilding services or waiting on a firewall change window.
- W-02
Policy you can review
Rules should read like a sentence and still compile to a deterministic verdict you can diff and explain.
- W-03
Agents with real reach
Agents call tools, hit egress domains and spend money. Limits have to be enforced, not documented.
How XAIGuard answers it
Three capabilities that carry the load.
Frameworks in scope
OWASP LLM01 Prompt InjectionOWASP LLM06 Excessive AgencyMITRE ATLASNIST AI RMF (MEASURE, MANAGE)
Next step