deadbits/vigil-llm
A Python library and REST API that scans LLM prompts and responses for prompt injections, jailbreaks, and other malicious inputs using multiple detection methods.
Velocity · 7d
+0.5
★ / day
Trend
→steady
star history
Vigil provides modular scanners including vector similarity, YARA rules, transformer models, and sentiment analysis to detect threats in LLM inputs. It supports local embeddings via SBERT or OpenAI, includes detection signatures and datasets for common attacks, and offers a Streamlit web UI for interactive testing. The tool is intended for security teams and researchers evaluating LLM pipelines.