Security engineers hardening LLM applications against prompt injection.

Vigil is a Python library and REST API that scans LLM prompts and responses for jailbreaks, prompt injection and other risky content.

Self-hosted, run locally as a Python library or deployed as a REST API service.

Created by researcher deadbits and still labeled alpha, with active issues and pull requests.

Provides a layered detection approach combining vector similarity, YARA rules, transformer classifiers and canary tokens.

Uses vector databases and transformer models for scanning and is meant to be one layer in a broader defense strategy.

Apache 2.0 open source and fully self-hosted with no external calls required, explicitly marked experimental and not a sole line of defense.