Tip Of AI
Discover what you can do with AI
AgentFox is an open source control plane for governing and securing AI agents across development and production environments. It scans repositories, configurations, and supported agent platforms to identify every agent, connected tool, and MCP server. The platform supports LangGraph, LangChain, OpenAI Agents SDK, CrewAI, AutoGen, Google ADK, MCP, and LiteLLM. AgentFox automatically drafts access policies based on the tools used by each agent, allowing teams to review and approve rules instead of writing them manually. Its runtime controls inspect tool calls, tool arguments, and the source of requested actions. Teams can begin in watch mode, where activity is recorded without blocking users, then replay policies against historical conversations before enabling enforcement. Risky actions can require human approval, with each approval limited to one specific call. Temporary exceptions can expire automatically. The product includes repository scanning, agent and tool discovery, runtime guardrails, audit trails, signed exports, policy replay, GitHub integration, and command line setup through uvx agentfox scan. In benchmark testing, AgentFox blocked 11.5 percent of harmless attack styled messages compared with 42.2 percent for an LLM Guard model, while catching 68.3 percent of real attacks compared with 31.7 percent. AgentFox is released under the Apache 2.0 license.