Zoona AIJuly 25, 2025
Train luna with documents and URLs to further refine its performance
Luna is SparrowDesk native artificial intelligence. It’s powerful enough to function as a standalone AI agent, answering customer questions and resolving problems end to end. And also as an AI Copilot, providing human agents with instant conversation summaries, ready-to-send reply drafts, on-demand knowledge bites, and anything else they need for quick and helpful support.
For optimum performance and personalized responses, Luna has to be trained with information. The more context you provide Luna, the faster and sharper its responses become.
By default, Luna learns from your company’s knowledge base and support interactions. Starting now, you can feed it with information from documents and URLs, so that it can acquire knowledge from your internal resources, like agent guidebooks, and external sources. This gives Luna an in-depth understanding of your business and industry while you have the flexibility to train it granularly without making the information public.
- Included new options such as timezone, page URL, country, language, and UTM parameters for efficient filtering and prioritization of conversations.
- Added a column in the members page to show who has Copilot Seat access, making team roles clearer.
- Introduced separate notification sounds for new incoming conversations and follow ups for better distinction, and optimized volume levels for better audibility.
- Made the chat widget’s close button more prominent for easy collapse and reopen.
- Corrected password policy enforcement to ensure lockout periods are applied exactly as configured, preventing unintended default lockout durations.
- Invalidated all active user sessions immediately after a password change to eliminate the risk of continued unauthorized access.
- Mitigated an IDOR vulnerability that could allow deletion of contacts across accounts by tightening account-level access validation.
- Fixed an issue preventing uploaded Luna agent avatars from rendering correctly as display images.
Stabilized the “Forgot Password” flow to prevent intermittent failures and ensure reliable account recovery.
