We need to edit production files. The team is pushing Cursor hard, but I'm not letting an AI IDE near our core code without a fight.

Two alternatives: Aider (terminal-based, git-aware) and Goose (local, open-source model runner). The safety question isn't about which has fewer CVEs. It's about data movement and trust boundaries.

Aider uses OpenAI's API by default. That's a hard no for proprietary code. You can point it at a local LLM, but then you're managing that stack. Goose runs everything locally, models and all. No data leaves the machine.

But "safer" depends on your threat model. If the risk is accidental code leakage, Goose wins. If the risk is a faulty edit causing an outage, Aider's tight git integration might actually be safer for rollbacks. Goose is just an editor.

I'm leaning towards the local-only approach. The compliance overhead of vetting Cursor's backend is massive. Goose's model is a known quantity you can airgap. What's the real ROI on letting an AI see the entire codebase versus just the diff you're working on?