AI Overview

Overview

Three switches on Settings, under AI, control what you can do.

AI Inference turns on Query Mode for an Analyst or an Admin. Marcus Aurelius answers questions about your assets, risks, and technologies. Query Mode does not run scans or change data. A Read-only user does not see the AI Assistant, even when AI Inference is on.

AI Agents turns on Agent Mode in Marcus. It also needs AI Inference. Only an Analyst or an Admin can use Agent Mode. In Agent Mode, Aurelius can delegate work to specialist agents. How that works is on Agent Mode.

AI Autonomy turns on Hannibal. Only an Analyst or an Admin can open a hunt. A Read-only user cannot.

Architecture

The AI system consists of three key components:

1. Conversation Engine — Manages stateful chat sessions between users and the AI, persisting messages in DynamoDB with full audit trails
2. Agent Framework — A tool-calling loop powered by Guard's hosted models that allows agents to query data, execute capabilities, and spawn sub-agents
3. RAG Pipeline — Amazon Titan embeddings with PostgreSQL pgvector for grounding AI responses in your organization's actual findings and knowledge base

Security and Isolation

All AI operations enforce strict multi-tenant isolation. Every query, agent execution, and conversation is scoped to your organization.