Project Guardian Released — A Firewall for AI Agents

🛡 Project Guardian Released — A Firewall for AI Agents

Project Guardian (v0.1.0) is a local firewall for autonomous AI agents operating in user-space. The system intercepts agent actions (files, shell, network) at the tool interface boundary or via an MCP gateway, using a deterministic policy mechanism (allow/ask/deny). Decisions are made by a predictable rules engine rather than an LLM, which protects against prompt injection.

🌍 A new approach to agentic system security is proposed, shifting control from the OS kernel to the tool-call boundary. This simplifies the implementation of protection and makes it independent of the specific AI model.

👤 It allows for the safe delegation of tasks to AI agents by providing an interface to approve their actions in real time.

Source 1: https://github.com/Vadale/project-guardian