GPT-6 Sol and Luna, fixes found by Sensei, and more reliable insights and embeds

Agent design in Perspective now uses GPT-6 Sol by default. Existing design sessions keep the model they were created with, but you can switch them to Sol in the model picker. GPT-6 Luna is also available there.

Sensei, our AI conversation quality coach, found each problem below by reviewing real conversations and production signals.

• When someone shares context before a tappable question, the agent now responds to what they said before presenting the choices. Quick replies feel like the next turn in a conversation, not a button panel that ignored them.
• Form Data no longer fills unanswered fields with answers nobody gave. Missing text, numbers, and yes-or-no answers stay empty instead of becoming an empty string, 0, or No. A conversation that ends before qualification is decided is no longer filed as disqualified.
• On the Team Members page, actions your role does not allow now explain why and keep the dialog open instead of looking like a server failure.

• Daily and weekly insights now spread their work across a short window, preventing the morning traffic spike that could temporarily make requests fail. Digests and invites still run at the exact time you chose.
• Embedded agents now ignore unrelated messages from other scripts on the page instead of throwing browser errors on unexpected data.