Ask Observer
Plain questions in.
Deterministic answers out.
A natural-language bar lives in the topbar of every dashboard screen. Behind it sits a deterministic grammar interpreter — unit-tested, with 13 intents and natural date ranges — that turns your question into the live dashboard query you meant. It is not an LLM, and that is not a limitation: it is the feature.
The askbar
One bar, every screen.
Press ⌘K or click the topbar anywhere in the dashboard. Type a question; results render as tables and cards in a modal — the same reports you already know, reached in one step.
Thirteen intents
What you can ask, exactly.
Every supported intent is enumerated below — the grammar is unit-tested against this list. Date ranges come naturally: today, yesterday, 7d, 28d, 90d, or plain phrases like “last 14 days”. Outside the grammar, Observer says so and shows suggestions — it never improvises.
| You ask | What Observer runs |
|---|---|
| top pages 7d | Pages report for the last 7 days |
| sessions with rage | Sessions list pre-flagged with rage clicks |
| multitab sessions | Sessions carrying the multi-tab badge (⎇N) |
| vitals /pricing | Core Web Vitals percentiles for /pricing |
| heatmap rage /pricing | Rage-click layer of the /pricing heatmap |
| deploy diff | Since-deploy deltas vs. an equal window before |
| anomalies | z-score anomaly cards across all metrics |
| frustration | Frustration leaderboard, worst pages first |
| forms | Form drop-off analytics across sites' forms |
| acquisition 28d | Channels and UTM breakdown over 28 days |
| realtime | Active visitors in the current 5-minute window |
| sites | The multi-site registry with per-site status |
| session 91be…40 | Full timeline for that session, per-tab lanes |
Why deterministic
No fake AI.
A grammar, tested.
Ask Observer is a deterministic grammar interpreter with a unit test suite behind it — the same questions parse to the same queries on every release. There is no language model between you and your data, so there is nothing to hallucinate with. When a question falls outside the 13 intents, the bar says so and offers what it can do instead.
Same question, same answer
Parsing is pure and deterministic — the interpreter is covered by automated tests that run on every release, not by a prompt.
Results are your reports
Every answer is a live dashboard query rendered as tables and cards in a modal — with the same exact / estimated / suppressed badges everywhere else.
Failure is honest
Ask something outside the grammar and Observer names the supported intents instead of inventing a plausible-looking number.
Ask it twice.
Get the same answer both times.
The askbar ships on every dashboard screen, in every tier — including the default one with no consent banner.