Conduct AI offers runtime governance for AI agents through two products: Guard, a policy engine that blocks, warns, audits, or injects actions before execution, and Router, an LLM proxy that routes requests through Guard. Both share one policy and repo, with signed configuration and hash-chained audit logs for cryptographic proof.

Key features include runtime firewalls that control actions before they happen, unlike observability tools that only report after. The three-pillar moat ensures config integrity via workspace-signed policies, audit integrity via SHA-256 chain, and coverage for LLM and shell/MCP calls. Failure modes are fail-closed by default.

Getting started is easy: clone the repo, run docker compose, and point your SDK at the local API. For CLI tools, install conduct-cli and log in. The platform ships with 20+ compliance packs (OWASP, SOC 2, HIPAA, PCI DSS, EU AI Act, NIST, ISO 42001) and 22 pre-built playbooks for common workflows.

Architecture includes a FastAPI backend, Postgres for state, Redis for queues, and a Next.js Canvas UI. Guard checks fire at three chokepoints: CLI hooks for coding agents, MCP layer for tool calls, and Router for any SDK. Deployment is self-hosted via docker compose.