I was using Claude Code to clean up some stashed changes. It ran git stash drop. No confirmation. Irreversible. Gone.
Not catastrophic, but the moment made me realize: I had no idea what else the agent had done that I hadn't noticed.
The problem
Most developers using Claude Code, Cursor, or similar tools have zero visibility into what these agents are actually doing. Every file read, every bash command, every API call — it happens and disappears. No log. No policy. No way to review it after the fact.
That's fine for a toy project. It's a real problem when something matters.
And it's a much bigger problem at the org level. Security teams are being asked to sign off on AI agent usage with nothing to audit and no way to enforce policy. They're just hoping the agents behave.
What I built
I wanted a layer that sat between me and the agent's tool calls. Something that logs every action with full context, enforces simple policies (don't touch these files, ask before running destructive commands), and runs entirely locally — no cloud, no code leaving your machine.
That's Oculi. A CLI that hooks into your AI coding agent at the IDE layer, intercepts tool calls before they execute, and checks them against policies you define in YAML.
For security leaders, there's a centralized management layer on top of that — think EDR, but for AI agents. One place to define and push policies across every endpoint in your environment, with a unified audit trail of every action every agent has taken across your org. Visibility and control at the fleet level, not just per developer.
If you're dealing with this — as a developer or on a security/platform team — waitlist is open at oculisecurity.com.
The gap is opening faster than most teams realize.
Top comments (1)
totally get the concern about losing work without a trace. that lack of visibility can be a real pitfall for developers. at moonshift, we focus on giving you a full next.js + postgres + auth app up and running in about 7 minutes, and you own the code on your github. if you're curious, i can offer you a free run to see how it goes.