Vigil is an open-source Python library and REST API that scores prompts and responses of large language models (LLMs) for prompt injection, jailbreaks and other risky inputs. Each configured scanner contributes to the verdict: vector-database similarity, YARA heuristics, a transformer model, prompt-response similarity, and canary tokens for detecting prompt leakage. You can add your own detections through YARA signatures. The project also ships signatures and datasets for self-hosting, plus a Streamlit web interface for experimenting. Vigil runs as an API server or is imported as a library into a Python application.
No company stands behind Vigil and there is no hosted service: you run it on your own infrastructure, so no data goes to a vendor. That holds only while you use local embeddings (SBERT); with OpenAI embeddings, text leaves your environment. The code is licensed under Apache-2.0. According to its maintainers the project is in an alpha state and should be treated as experimental or a research tool, so a European organisation using it in production has to take care of maintenance itself.