Changelog

AI agents now judge your escalation rules more reliably. They tell real requests for a human apart from passing mentions, and anger that has already been resolved no longer triggers a handoff.

Customer email replies to your outgoing emails now stay in the same conversation instead of sometimes starting a new one, so your team and AI agents keep the full context.

The cursor goes straight into the reply box when you open a conversation, and Shift+Enter now sends your reply. Enter still starts a new line.

Only workspace admins can now create and manage API tokens and webhooks. The API token list also shows who created each token. Admins may want to check their existing webhooks.

The "Assigned to the team" notice in the chat widget no longer disappears when a customer reloads the page after their chat is handed to your team.

If the chat window fails to load, the widget tries again automatically. If it still can't load, visitors see a clear message with "Try again" and "Close" buttons instead of an empty panel.

Signing in with an email that isn't linked to a workspace no longer creates a new workspace by accident. You'll see a clear message on the sign-in page with your email kept in place so you can correct a typo or register if you're new.

Closing a conversation right after adding a tag no longer reopens it, and tag-filtered custom view counts now update immediately instead of requiring a full page refresh.

Shared report views and customer segments now appear under the correct section of the sidebar rather than being listed as private.

Reopening a saved view filtered by assignee or time now shows its filters correctly instead of appearing empty.

Agent-level reports now show figures for that agent only; previously they could fall back to organisation-wide numbers.

Bulk-moving training items to the top-level "Content" folder now works as expected.

Documents that fail to process (e.g. an unreadable PDF) are now clearly marked as failed rather than having error text quietly added to your agent's training data.

The People tab in the company details panel now shows your real customers — placeholder sample contacts no longer appear when no data is available.

Fixed an issue where some users stopped receiving their scheduled customer-leads digest emails, and corrected API error responses when creating agent memories or updating training data with an invalid agent.

The inbox recovers automatically after a dropped connection or a laptop waking from sleep. No more stuck "Error loading conversation" pane; the list and open conversation refresh on their own when you're back online.

Creating, editing or deleting tags and saved views now gives clear feedback when something goes wrong (such as a duplicate tag name). You'll see a single, specific error message and can fix it without losing your place.

Your AI agent now checks your knowledge base on every customer question, and prioritizes authoritative sources; Q&As, uploaded files and your own website pages rank above blog and news content. Answers are more accurate and grounded in the material you've provided.

Real-time delivery is more reliable: very long messages (for example, emails with large bodies) now appear in the inbox instantly instead of occasionally failing to show until a refresh.

Released an enhance version of our search tool making it easier to find conversations, training data or reporting by keyword