For LLM applications hosted on any infrastructure:
Dome blocks unwanted inputs and outputs in real-time
Driven by policies that you can specify in natural language, Vijil Dome implements guardrails that block adversarial prompts, prompt injections, jailbreaks, PII leakage, toxic content, and other policy violations.
Policy-Driven
Upload your organization policies (for example, agent instructions, company code of conduct, regulatory requirements, and industry standards). Vijil Dome translates your intentions into guardrails to enhance security and safety, automatically constructing a custom perimeter defense around any LLM you use.
Comprehensive
Covers not only well-known threats including security, privacy, and toxicity, but also more pernicious vulnerabilities such as ethical violations against the company code of conduct or EEOC regulations against bias in the workplace.
Accurate
Up to 2x more accurate than Amazon Bedrock guardrails and 95% of human evaluation, Vijil Dome guardrails use fine-tuned LLMs, custom ML models based on embedding vectors, and pattern-matching algorithms to detect out-of-policy errors with high-precision.
Transparent
Vijil Dome captures and stores detailed logs of all inputs and outputs, and all guardrail decisions, enabling analysis and visualization in your preferred platform.
Adaptive
Learns from logs and traces with privacy-preserving machine learning models to tune itself to the particular usage patterns of your users in your deployment.
Fast
Loads as a Python library or Docker container inside your application, wrapping around API calls to the LLM seamlessly with minimal latency of less than 300 ms.
Try
Dome for FREE
Get started with Vijil Dome at no cost for 30 days. Contact us for details.