Developers using AI coding assistants face a hidden security risk: tools like Cursor and Claude Code often embed API keys in local chat history databases that standard scanners miss. These secrets persist unencrypted in SQLite files outside version control, creating potential exposure points.

Sieve addresses this gap by scanning transcript stores from Cursor, Claude Code, Copilot, Windsurf, and other popular assistants. The tool identifies exposed credentials and offers in-place redaction directly from VS Code's .vscdb files. Built specifically for macOS, it handles the unique file structures where AI tools store conversation history.

Priced at $9.99, Sieve includes a vault feature leveraging macOS Keychain for secure secret storage. The application uses security-scoped bookmarks for folder permissions and ships with an open-source core called SieveCore. All scanning happens locally with zero network requests or telemetry collection.

This tool fills a genuine security void for developers increasingly reliant on AI assistants. By catching leaked credentials before they cause damage, Sieve provides peace of mind for teams serious about secrets hygiene.