Voice, Listening, and Persistence

Voice behavior

Voice is not a presentation layer added after the product works. It is the primary operating surface for live work: the user speaks, the system reasons with context, and the response streams back while the decision is still active.

The experience supports speech input, spoken response, and streaming conversational output. Text remains an important fallback so the system still works if voice is unavailable.

Persistence

• Resume threads later
• Recall prior decisions and facts
• Continue background work after the user pauses
• Surface relevant context when the user returns
• Carry approval and action state forward instead of treating each turn as isolated

What users should expect

• Voice is best for live interaction
• The system may ask clarifying questions before modeling an outcome
• Sensitive actions may pause for approval
• Background tasks may continue after the conversation pauses
• Voice should have a text fallback when the environment is noisy or unavailable

Reviewer caveat

Public docs should describe the intended product behavior and current product surface without implying that every deployment, audio path, or customer environment has already been independently accepted. Browser-visible and browser-audible proof should be verified separately for any live demo claim.