HeimWall

HeimWall is a Mac menu bar app that watches what you're about to send to AI coding assistants and flags secrets or personal data before they leave your machine. It works across Cursor, Claude Code, and Copilot, scanning your prompts for credentials like AWS keys, GitHub tokens, private keys, JWTs, and database URLs. It also catches personally identifiable information like social security numbers, credit card numbers, and email addresses.

The problem it solves is one most developers have accidentally created at least once. You copy a config file or log snippet to ask an AI for help, and buried in there is a production API key or a customer email address. By the time you realize it, the data has already been sent to a cloud service. HeimWall intercepts that moment, running detection locally before anything transmits.

Everything happens on your Mac. There's no server to phone home to, no account to create, and no data collection. The detection uses 25-plus hand-written regex rules for common credential patterns combined with a pretrained on-device classifier as a fallback. Scanning takes under 50 milliseconds, so it doesn't slow down your workflow. The app runs under 150 MB of RAM while idle and weighs about 15 MB to install.

The target audience is individual developers using AI assistants who want a safety net against accidental leaks. If you work with production credentials regularly and use Claude Code or Cursor as part of your daily flow, this sits quietly in your menu bar and only interrupts when it spots something risky. It's also useful for anyone whose company cares about data handling but hasn't rolled out enterprise-grade DLP tooling.

HeimWall is free forever. No subscription, no tiered plans, no account signup. You download it, install it, and it runs. The developer explicitly states that data is never collected or used for performance reviews. The app is signed and notarized by Apple.

Current availability is Apple Silicon Macs running macOS 13 or later. Intel Mac and Windows versions are planned, with a waitlist for those interested. The latest version at time of writing is 0.0.5.

Where HeimWall differs from cloud-based security tools is the strict local-only approach. There's no compromise where metadata goes to a server for analysis. If privacy is the concern, the architecture matches the promise. The tradeoff is that updates to detection rules require an app update rather than a backend push, but for a tool this simple, that's reasonable.