G.A.I.N. 0.5.22 improves live browser enforcement so redaction follows the policies configured in your dashboard. Mixed prompts now redact only the data types covered by active policies, including API keys, secrets, and database connection strings, while leaving unrelated text unchanged.
Also includes improved regression coverage for crypto seed phrases and mixed sensitive-data prompts.

G.A.I.N 0.5.20 improves reliability and performance for AI prompt and file protection:

• Reduces per-tab memory work by lazy-loading seed-phrase word data only when needed.
• Caches send-button candidates to avoid repeated full-page queries while scanning.
• Hardens multi-file upload handling so mixed warn/redact/block batches do not bypass warnings.
• Fails closed when PDF/DOCX text extraction fails, blocking unscannable sensitive uploads instead of allowing them silently.
• Adds regression coverage for file-upload, redaction, policy, and network-guard cases.

Initial release of G.A.I.N.

G.A.I.N runs sensitive-data detection locally in your browser on supported AI tools
(ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, DeepSeek). When it detects
secrets or personal data in a prompt or an attached text file, it applies your
organization's policy: log, warn, redact on the device, or block before the data is sent.

Only metadata (the AI tool used, detector category, severity, and action) is sent to the
organization dashboard. Prompt text, pasted values, and file contents never leave the device.

G.A.I.N 0.5.4 adds improved organization onboarding, department selection during setup, live device presence, managed deployment support, and stronger local redaction behavior across supported AI tools.

This version also improves policy sync, uninstall handling, and metadata-only reporting for team deployments.