vigil-models

Vigil Models

Fine-tuned safety models for the Vigil AI agent safety framework.

Models

vigil-v1 (4.4 GB)

• Base: Qwen 2.5 7B (Q4_K_M quantization)
• Purpose: Full safety classification model — prompt injection, jailbreaks, data exfiltration, unsafe tool calls
• Usage: ollama pull hexitlabs/vigil-v1 (if published to Ollama registry) or download GGUF below

vigil-lite (986 MB)

• Base: Qwen 2.5 1.5B (Q4_K_M quantization)
• Purpose: Lightweight safety model for edge/CPU deployment — 5x smaller, 2.5x faster, 93.3% accuracy
• Usage: ollama pull hexitlabs/vigil-lite (if published to Ollama registry) or download GGUF below

Files

• vigil-v1.gguf — Full 7B model weights
• vigil-lite.gguf — Lite 1.5B model weights
• Modelfile-v1 — Ollama Modelfile for vigil-v1
• Modelfile-lite — Ollama Modelfile for vigil-lite

Quick Start (Ollama)

# Download the GGUF file, then:
ollama create vigil-v1 -f Modelfile-v1
ollama create vigil-lite -f Modelfile-lite

Built by HexIT Labs