LOCKE FOR CURSOR
Get whole-codebase answers from Cursor without shipping the whole codebase.
Cursor goes a step beyond an autocomplete plugin: it's an editor built around the model, designed to reason across your entire project. Ask it to refactor a pattern, explain a subsystem, or trace a bug, and it draws on a broad index of your codebase rather than just the file in front of you. The breadth is the feature — and it widens the surface of what can be sent to a model in a single interaction.
Project-wide retrieval means a question about one module can sweep in context from files you weren't looking at — the config with live credentials, the seed script with real user data, the internal doc pasted into a comment. The developer's mental model is "I'm asking about my code," while the system's behaviour is "I'm assembling context from across the repo." That mismatch is where sensitive material slips through.
Locke sits at the boundary between your machine and the model. It identifies secrets and regulated data in the context Cursor gathers and masks them before they're transmitted, so you keep the editor's project-wide intelligence without exposing the project-wide secrets that come with it.
The risk with Cursor
Cursor indexes and retrieves across your whole project, so a single question can sweep in credentials, secrets, or real customer data from files you weren't even looking at — far beyond what you'd ever knowingly paste.
How Locke helps
Locke identifies sensitive data in the context Cursor assembles and masks it before it reaches the model. Locke — the Sonomos desktop app, coming soon — runs locally between Cursor and the model so the protection follows your editor, not just a browser session.
Keep using Cursor — without the exposure
Locke runs entirely on your device. Sensitive data is detected and masked before any prompt is sent, so nothing confidential ever leaves your machine. Pricing for Locke, the desktop app, is coming soon; Canary is free and open source.
