Unplug AI

About

Unplug provides runtime security for agents in LLM applications. It monitors the origin of text, whether from the user, retrieved data, or tool outputs, and checks for prompt injection attempts and harmful actions. The system also enforces policies on tool calls, using precise span-level redaction rather than blocking entire messages. While typical guardrails often discard complete messages, Unplug identifies the specific segment of an attack and excises it precisely, allowing the remainder to stay functional.