A few quiet but meaningful changes to how the AI handles conversations:
It remembers what it did
If the assistant ran a tool earlier in the conversation — created a ticket, forwarded to a teammate, looked something up — it now knows that the next time you ask. "Did you do it?" gets a real answer instead of a generic one.
It recognizes returning customers
When someone has reached out before, the AI sees a short summary of their history (prior threads, open tickets, last contact) and frames its reply accordingly. It can also pull full ticket history on demand if the customer references a past issue.
No more duplicate tickets
Under the hood, every action the AI takes is now idempotent. If a webhook is delivered twice, or the system retries after a hiccup, you'll still get exactly one ticket, one notification, one client email. Whatever was supposed to happen, happens once.
You decide whether sources show
New toggle in AI Settings: keep knowledge base citations visible in replies, or hide them entirely. When hidden, the AI speaks as inherent expertise and any stray article link gets stripped before send.
WhatsApp replies look right
Markdown now gets converted to WhatsApp's native formatting (bold, italic, lists), and the AI knows it's on WhatsApp — no more "click the button below" when there is no button.
Nothing to do on your end. These are live for everyone.