AgentGuard

AgentGuard is runtime governance software for production AI agents. It is built for APRA-regulated financial services in Australia and APAC, and it focuses on giving organizations CPS 230 evidence from day one, with EU AI Act and ISO 42001 mappings included.

The system enforces policies inside the agent runtime rather than by scanning prompts. It intercepts tool calls, model outputs, and dollar-spend before they happen, and its YAML policies can block actions, require human approval, or escalate work to a second agent. The policy examples shown on the site include tool whitelists, dollar caps, PII egress controls, human-in-the-loop gates, and after-hours rules. AgentGuard also produces signed, Bitcoin-anchored audit records for agent decisions, and it assembles those records into a single PDF compliance pack with framework mappings, control owners, and statistical sampling.

AgentGuard is shown as governing OpenClaw, LangChain, CrewAI, AutoGen, OpenAI Assistants, LangGraph, MCP, Vercel AI, and Express/FastAPI. The site says it installs with one line in Python or Node, works locally and behind a firewall, and has no SaaS dependency for runtime. It provides a decorator or middleware per framework, and for OpenClaw it is described as a drop-in plugin with per-claw policy, fleet view, and a killswitch and rollback on policy breach.

Pricing includes a Free plan at $0 forever for one agent with 7-day retention, local audit log, Community Discord, and an MIT-licensed SDK. The Team plan is $499 AUD per month for up to 25 agents, with 90-day retention, SSO for Google and Microsoft, signed audit roots, and email support. The Compliance plan is $2,500 AUD per month flat for unlimited agents, 1-year retention plus cold archive, CPS 230 evidence pack, EU AI Act and ISO 42001 mappings, Bitcoin-anchored audit roots, and a quarterly readiness review. An Enterprise option adds on-prem deployment, an SLA, and a dedicated SE. It is developed by The Bot Club Pty Ltd.