Reading the daily brief
The brief has six sections: needs attention, today, waiting for, you promised, at risk, and whatever is empty says it's empty. It's capped at 250 words so it stays readable in under thirty seconds.
It's generated from structured queries against your own records, so the numbers in it are the numbers in your database — you can tap through and count them.
When the brief arrives
The brief is generated on a schedule for your workspace. Per-user, timezone-aware send times are a tuning item rather than a shipped feature, so the timing is currently the same for everyone in a workspace.
What the assistant can look up
It has read tools for: your actions, overdue actions, what's due today, waiting items, everything involving a person, everything on a project, recent decisions, open commitments, risks, and a full-text search across your work.
Every fact in an answer comes from one of those calls. When they return nothing, the answer is that there's nothing recorded — not a plausible guess.
Questions that work well
- "What am I waiting for?"
- "What did I promise Sarah?"
- "What's overdue?"
- "What's happening on the Northgate project?"
- "What did we decide about the supplier?"
Questions that won't work are ones about things you never captured, and general knowledge questions — it's an interface to your work, not a chatbot.
What the assistant cannot do
It has no external tools. It cannot send an email, book a meeting, message anyone or take any action outside Zephyr. This isn't a permission that's switched off — the capability doesn't exist.
It can propose creating or changing an action, which you then confirm in the UI.
Conversations don't persist yet
Each assistant session starts clean. A follow-up question that refers back to something said earlier in a previous session won't find it. Persisted threads are planned work.