More accurate analysis counts, plus fixes for conversation recovery, voice endings, and automations

Sensei, our AI conversation quality coach, found these problems by reviewing completed conversations and production signals.

• Empty conversations opened through AI assistants no longer leave you charged. After inactivity, their credits are returned and they are hidden from Results.
• Fixed cases where repeating a surname replaced the participant's saved full name in Results.
• Fixed cases where browser conversations ignored outline rules based on the participant's device, browser, or approximate location.
• Fixed cases where an unreadable attachment, such as a password-protected PDF, left text conversations unable to reply.
• Fixed more cases where text interviews ended before the remaining questions were asked.
• Fixed cases where a Live voice call stayed silent after the agent said it was wrapping up.
• Reduced extra narration before the final goodbye in voice conversations.
• Fixed counts in analysis sessions and daily insights that included owner previews or grouped separate anonymous conversations as one participant.
• On newly created agents, weekly insights are scheduled later than daily updates so they can include that day's findings.
• Unsupported webhook automation changes now show an error instead of saving settings that prevent delivery.
• Design chat now marks interrupted documentation answers as incomplete, so you can see when a lookup did not finish.

• Fixed cases where Analysis chat showed an error page while a report was still being written.
• Fixed cases where a failed text reply left a blank message in Results.

Behind-the-scenes security updates to keep your workspace safe.