Hey everyone! I'm Lokesh, the creator of ZeroTrust Agents. Over the last few months, watching the explosion of autonomous AI agents (LangChain, CrewAI, AutoGen) has been incredible. But as engineers, it also terrified me. We are giving probabilistic LLMs the ability to execute SQL, hit Stripe APIs, and write to file systems. Right now, developers are trying to secure these agents using "prompt engineering" (telling the LLM "please don't delete my database"). But as we know, prompt injections easily bypass this. Security shouldn't be probabilistic. That’s why I built ZeroTrust Agents. It’s an open-source, deterministic API Gateway that sits between your Agent and the LLM (OpenAI, Anthropic, Gemini, MCP). How it works: Interception: We proxy the stream and buffer any attempts the LLM makes to call a tool. Policy-as-Code: You write YAML rules to allow/deny tool execution. Human-in-the-Loop: For high-risk actions (like processing a refund), the Gateway suspends the agent's HTTP request and pings you on a dashboard to click "Approve" before the tool runs. You don't need to rewrite your agents. Just change your SDK's base_url to point to the firewall, and you're secure. We are completely open-source. I would love for you to check out the GitHub repo, try running it locally via Docker, and let me know your thoughts. I’ll be here all day answering questions! What tools are your agents using that you wish you had a firewall for?

ZeroTrust Agents

A zero-trust API Gateway for your AI Agents.

A zero-trust API Gateway for your AI Agents.