👋 Maker here. I built Nimuvo because I kept catching myself using AI coding agents more and more, and getting less sure of what they were actually touching. It's not that I think the agents are malicious — it's that one misunderstood instruction with broad permissions can do a lot of damage before you notice. The first version of this was me hand-editing Claude Code hooks, a half-broken Git pre-commit guard, and a sticky note that said "don't let it touch .env." That worked for about a week. So I packaged it. **What Nimuvo is right now** (honest scope, no vapor): - Works with Claude Code, Gemini CLI, local Cursor Agent, and Codex - Folder boundaries + reviewed destructive-command rules - Local Ask workflow for risky actions (Deny / Allow Once / 15 min / Session) - Encrypted local activity, no telemetry - Free during launch, no account required **What it isn't yet:** - Not system-wide (no Endpoint Security yet — that's an Apple entitlement I haven't shipped) - Doesn't cover generic terminal commands or MCP tools - Cursor Agent is hard-block-only — no Ask yet, their hook surface is limited **What I'd genuinely love feedback on:** 1. Does the Ask UI give you enough context to decide in under 5 seconds? 2. What's the smallest rule set that would make you trust this on day one? 3. Which agent or workflow do you want protected next? Happy to answer anything in the comments. If you try it and something's broken or confusing, I'd rather hear it now than later.

Nimuvo — AI Safety Layer for macOS

Give your AI coding agent room to work, not your whole Mac.

Give your AI coding agent room to work, not your whole Mac.