We built a fully deterministic control layer for agents. Would love feedback. No pitch
Most of the current “AI security” stack seems focused on: • prompts • identities • outputs After an agent deleted a prod database on me a year ago. I saw the gap and started building. a control layer directly in the execution path between agents and tools. We are to market but I don’t want to spam yall with our company so I left it out. ⸻ What that actually means Every time an agent tries to take an action (API call, DB read, file access, etc.), we intercept it and decide in real time: • allow • block • require approval But the important part is how that decision is made. ⸻ A few things we’re doing differently Credential starvation (instead of trusting long-lived access) Agents don’t get broad, persistent credentials. They effectively operate with nothing by default, and access is gr...
