Custom LLM guardrails and observability
To address emerging regulatory requirements, enterprises need a solution for enforcing their bespoke governance policies to prevent misuse, while auditing LLM usage in real time. With DynamoGuard, compliance, legal, infosec, and product teams can copy-and-paste their AI governance policies into DynamoGuard. DynamoGuard will flag edge-cases for compliance teams to help fine-tune DynamoGuard to enforce your bespoke governance policies. Deploy DynamoGuard in real-time to moderate and flag non-compliant LLM interactions.
Customize Policies in Natural Language
Directly describe your internal AI policies in natural language within DynamoGuard to autogenerate guardrails customized for your business
Guardrail against Emerging Attacks
Defend against any type of risky input / output across PII leakage, toxic content, jailbreaking, prompt injection attacks. Identify / defend against any type of non-compliance as defined by your specific AI policies
End to End Observability
Capture live user interactions and generate real-time analytics / alerts to monitor non-compliant LLM interactions. Audit searchable user logs for comprehensive documentation.
.webp)
Minimal Latency & Compute Overhead
DynamoGuard offers best-in-class guardrails that both minimize latency overhead and cost, while beating LlamaGuard's F1-score in areas such as toxicity detection.
How to deploy DynamoGuard?
The DynamoGuard integrates seamlessly with a variety of deployment scenarios. Select the one that best fits your organization
Cloud can be deployed on:
- Base LLM in Cloud
- Guard Agent in Customer VPC
- Guard Platform in Customer VPC
Requires internet connection LLM communication is ongoing between DynamoGuard in Lenovo VPC and Base LLM in Cloud
Hybrid (browser extension or application) can be deployed on:
- Base LLM in Cloud
- Guard Agent on Device*
- Guard Platform in Lenovo/Customer VPC
Requires internet connection LLM communication is ongoing between Guard Agent and Guard Platform Updates can be pushed via Device Manager or VDI
On-Device (browser extension or application) can be deployed on:
- Base LLM on Device
- Guard Agent on Device*
- Guard Platform in Lenovo/Customer VPC
Can run with or without internet connection when connected to the internet, LLM communication is ongoing between DynamoGuard Agent and DynamoGuard Platform. When not connected to the internet, DynamoGuard acts based on recent push Updates can be pushed via Device Manager
DynamoGuard Lite On-Device (application) can be deployed on:
- Base LLM on Device or in Cloud
- Guard Agent on Device*
- Guard Platform N/A
Can run with our without internet connection. LLM communication is pre-packaged with limited default policies.
