Under every chapter of your brief sits a small label: where that material came from. Something you added, like a feed or your notes, and the label names it. The model's own knowledge, and the label says so and dates it. It is one line of gray text, and it is the most important thing we build.
A voice is a trust machine. When a warm human-sounding voice tells you something over breakfast, your defenses are down; that is precisely what makes audio briefings wonderful and precisely what makes them dangerous. The medium removes the visual cues, the URL, the byline, the design of a source you would use to calibrate on a screen.
Three rules the composer cannot break
First, no invented breaking events. The composition engine is contractually forbidden, in the literal sense that the output contract rejects it, from framing its own general knowledge as this morning's news. If your account has no live source for a topic, your brief speaks in honest tenses: trends, context, what to watch, never 'this just happened'.
Second, every chapter declares its basis in structured data, not prose. The label you see is rendered from the same field the system validates, so a chapter without a basis cannot exist in our database at all.
Third, on contested topics the brief describes the disagreement and attributes the positions. It briefs you; it does not recruit you. A personal show that quietly optimized for your agreement would be a radicalization engine with good production values, and we are not building one.
Boring on purpose
None of this demos well. Labels and refusals never do. But a daily habit product lives or dies on the morning you catch it being wrong, and we would rather you catch it being honest. The gray label is our signature on every item, and we think in a few years every serious AI media product will have one.
Give your website a voice
Free plan, grounded answers with sources, honest handoff. No card.
Build my agent