Feature Discovery Nudges
This feature is retired. Feature Discovery Nudges — the roughly-daily inbox card that pitched one feature you had never used — is off and hidden for every user as of 2026-08-31, and there is no toggle that turns it on. This page records what it was, how a nudge opened and behaved, and how its effectiveness was measured.
On this page
What it is
Retired 2026-08-31 (owner decision): nudges are off for ALL users. The registry entry
feature-discovery-nudgecarriesretired: trueand the display name "Feature Discovery Nudges (retired)". Retirement is stronger than off-by-default: the persistedfeatureDiscoveryNudgeEnabledsetting is ignored entirely, the Settings → Lab toggle is dropped, and the inbox source stops rendering already-released rows. Nothing you can set in the app brings a nudge back. The code is kept and reversible, and the only reveal left is the dev-onlyAMC_SHOW_FEATURE_DISCOVERY_NUDGE=1environment variable, which exists for testing — not as a user setting.
The rest of this page describes the feature as it was built, for the record.
A gentle, roughly-daily inbox card that pitched one Omniscio feature you had never used — "Here's a feature you're missing, and here's why it'd help." Selecting the card opened its detail in the main panel: the feature's name, a short hand-written pitch, a clean animation, and two buttons: "Start using it" turned the feature on (if it was off) and took you straight to it, while "Maybe later" dismissed the card for good — the same as the X. Nudges rotated through the features you didn't use, released at most one per ~24 hours, auto-stopped once you'd seen them all, and each card was dismissible and snoozeable like any other inbox row.
Some nudges opened a full landing page instead of the short pitch card — see below.
It was off by default and opt-in: while it was live, the registry entry
feature-discovery-nudge carried status: 'in-development', so nobody saw a nudge until
they turned on Feature Discovery Nudges in Settings → Lab (which flipped
featureDiscoveryNudgeEnabled; AMC_SHOW_FEATURE_DISCOVERY_NUDGE=1 was the dev-only
reveal). That toggle no longer exists — the 2026-08-31 retirement dropped it, because every
install that had the toggle on during the earlier shipped window kept a persisted ON setting
and went on receiving nudges after the feature was switched back off. The built-in bounds
were the user controls: at most one card per ~24 hours, never the same card twice, features
you already use were skipped, every card was dismissible and snoozeable, and the rotation
auto-stopped once every card had been seen.
Where to find it
Nowhere in the app. With the feature retired, no nudge card, landing page, toggle or Stats tab is reachable on a normal install; the sections below describe how the surfaces were built.
Landing pages — the richer sell-and-teach detail
Some entries open a full landing page instead of the short pitch card: a proper sell-and-teach page with a headline, a self-running demonstration of the feature, why you'd want it, and how to use it. Entries without one keep the classic short pitch card.
41 features have one today — Add an Account, Agent Status Board, AI Suggestions, Alarms, Asides, Ask Omniscio, Auto Context, Auto-replies, auto-waiting, Cron Jobs, Daily Digest, Desktop Icon Badge, Distribute Sessions, Drip, Email Cleanup, FlowVoice, Focus Mode, Gmail Integration, Inbox, Isolated Sessions, Keyboard Shortcuts, mobile access, Nighty Tidy, Omni Briefing, Parallel Sessions, Plain Speak, Quick Launch, Quick Replies, Recipes, Scheduled Messages, Search, Self-Healing, Semantic Search, Sharing, Skills, Snooze, Stats & Records, Text-to-Speech, Usage Forecast, Voice Input, and Zap. These pages are published artifacts, not hand-coded screens: each one is a designed page rendered right inside the nudge, and its copy or animation can be refreshed by re-publishing the page — no app update needed. Most of them are deliberately first in the rotation, so a page-backed pitch is one you actually see rather than one queued months out.
Only an approved, officially handed-off page is ever wired to a nudge — a page that merely exists and renders is unfinished work and never reaches you.
Snooze is the clearest example. Its page shows a noisy miniature inbox: press H (or
right-click → Snooze), say when in plain English ("tomorrow 9am"), and watch the card leave —
and come back exactly on time. Its button is "Try it on this card": instead of navigating
anywhere, it opens the real snooze palette on the nudge card itself. Pick a time and the card
vanishes from your inbox, then returns when you said — the feature demonstrates itself on a
zero-risk target. (Trying it does not dismiss the nudge; the card's return completes the
lesson, and you can dismiss it then.)
Under the hood an entry points at its page by URL, and the pane renders that page as the whole body (catalog.ts, FeatureNudgePageEmbed.tsx). If a page can't be reached, the card falls back to its short written pitch rather than showing you an empty panel. The older hand-coded style — a tagline plus a lazy-loaded animated demo component, with a still storyboard when "reduce motion" is on — is still fully supported (FeatureNudgeLandingContent.tsx, SnoozeDemo.tsx), though no entry uses it at the moment.
Fixed alongside this feature: snoozing a feature-nudge card from the right-click menu (or
H) used to be a silent no-op — the snooze was recorded under the wrong key, so the card never hid. Nudge cards now snooze like every other inbox row, and the detail pane closes itself once its card is snoozed.
Inline on/off toggle nudges
Some nudges don't send you anywhere — they put a live on/off switch right in the card.
Instead of "Start using it", the card shows the feature's current state as a real toggle you
can flip either way, plus a "Done" button that just closes it (your choice is already
saved). The first one is the desktop icon badge: your app icon (Windows taskbar / macOS
dock) normally shows a little count of sessions that need you — this nudge lets you keep it
or turn it off without leaving the card. The same switch also lives in Settings →
Notifications ("App icon attention count"), and flipping it in either place updates the badge
immediately. Under the hood this is the catalog's toggleSettingKey start-action bound to a
boolean setting (read !== false so an existing install with no saved value defaults ON); the
nudge auto-stops once you've engaged the control.
How it behaves
How it works
- What it knows you've used comes from the local
feature_eventstelemetry — a curated feature with zero matching events is "never used" and eligible to pitch. Nothing leaves the device for this decision, and no question, message, reply, overlay text, or account detail is ever recorded: each row is a feature name, an action name, and a timestamp.- Most entries reuse telemetry that already existed. A few features had none at all, and a
card cannot say "you've never used this" without a way to tell a first-timer from a daily
user — so pitching those meant adding a usage signal first. Plain Speak records
a count when you deliberately switch a message to its Plain Speak view (the ⇄ pill or the
Vhotkey — switching back records nothing, and a view the app pre-selects for you is never counted as your choice), and Zap records a count when you send one. Ask Omniscio records a count when you actually ask something (either the Ask tab or "Ask about this page") — never the question itself. Distribute Sessions records a count the first time it genuinely spreads a session across two or more of your accounts — no account name, id, or email, and once per app run rather than per session.
- Most entries reuse telemetry that already existed. A few features had none at all, and a
card cannot say "you've never used this" without a way to tell a first-timer from a daily
user — so pitching those meant adding a usage signal first. Plain Speak records
a count when you deliberately switch a message to its Plain Speak view (the ⇄ pill or the
- What it can pitch is a hand-authored catalog (catalog.ts) — a curated set of notable features (including Snooze and a "Check in on your Stats & Records" card), each with a one-line pitch and a one-click start action (enable a toggle, navigate to the feature, flip a live on/off switch right in the card, or — for try-it-live entries like snooze — act right on the card). The catalog is the single source of nudge wording and landing-page content.
- The daily scanner (feature-nudge-scanner-service.ts) runs hourly: if the feature is visible (which, since the retirement, requires the dev-only env reveal) and the last nudge was ≥24h ago, it picks the first catalog entry you haven't used and haven't already been pitched, inserts a row in its own
feature_nudge_releasestable, and pushes it to the inbox. When nothing is left it simply stops. - The inbox card is a standalone inbox source (its own
feature_nudge_releasestable — it does NOT use the shared alert/Drip table). It groups under a "DISCOVER" header. Dismissing archives it; snoozing rides the universal inbox-snooze table; a dismissed feature is never pitched again. - The in-pane detail (FeatureNudgeDetailPane.tsx) renders in the main panel via the generic
activeInboxDetailroute (like recipe / pr-inbox rows), so the inbox cursor can land on it. "Start using it" flips the feature's setting on (if gated), marks the nudge accepted, and navigates you to the feature; if turning it on fails it shows an error instead of leaving you stranded. "Maybe later" archives the card (same as the X) AND advances the inbox to the next item — it routes through the one shared inbox-dismiss chokepoint (dismissActiveInboxItem) that every reading pane uses, which captures the visually-next card and moves you to it, exactly like archiving an SMS or a weekly summary. (It does NOT merely deselect: a plain deselect would leave nothing selected and the inbox would re-open the newest card, bouncing the panel straight back open on this same nudge.)
Seeing whether nudges are working — the Nudges tab
While Feature Discovery Nudges was live, a Nudges tab appeared in Stats (it stayed hidden otherwise, so it never cluttered Stats for people who didn't use nudges). The feature is retired, so the tab no longer appears on any install. Per nudge you'd been shown, it puts how much you used the pitched feature before the nudge next to how much after it — side by side, with the difference — plus where the nudge landed (shown → accepted / dismissed / still-pending) and a "still filling" marker while the after window hasn't fully elapsed.
The "before" figure is ~0 by design: a nudge only ever fires for a feature you've never used, so the honest signal is the after adoption and the accept/dismiss funnel — a non-zero "before" is a red flag that a nudge was mis-targeted, not a bug. The window is a fixed 14 days on either side of when the nudge was shown, anchored on the show date (so every nudge gets a row, not just accepted ones). An agent calling a feature through the CLI does not count as you using it. Raw usage events are only kept for about 90 days, so for a nudge shown long ago the tab says data expired for a side whose window is older than the kept history, instead of showing a misleading 0; the difference only appears when both sides are known.
This reads existing data — the feature_nudge_releases rows plus the same feature_events
telemetry the nudge system already uses — and adds no new tracking
(queries-nudge-effectiveness.ts). The same
per-nudge numbers, scrubbed to anonymous counts (no timestamps, no personal data), also ride
the daily fleet digest so the product team can see fleet-wide which nudges actually move people
to adopt a feature.
Seeing what WILL happen — the preview
The Nudges tab above looks backwards at nudges already shown. The preview is the look forwards, and it exists because nudge delivery is otherwise almost unobservable: at most one card every 24 hours, chosen in catalog order, and gated off by default — so "is this actually working?" could only be answered by waiting a day to see whether anything turned up.
GET /feature-nudge/preview (bearer read, like GET /feature-nudge)
It answers three questions at once:
- Would anything land at all right now?
canLand, plusblockedBynaming the specific gate when the answer is no —hints-off(Feature tips, the switch for this class, is off — it ships off),flag-off(the feature itself is switched off),paused(the scanner service is paused), orcadence(a card already went out; the next is not due yet).explanationsays the same thing in plain English. - Which card is next, and when?
next, plus the full projectedschedule— every remaining card, one per 24 hours, in the order they will actually fire. - Why is a card NOT in the list?
skippedgives a reason per entry:already-pitched(spent — a nudge is never re-pitched, even after you archive it),already-used(you already use that feature, so there is nothing to discover), orineligible.
Two honesty details worth knowing. Each schedule entry is a window (earliestAt to
noLaterThan), not a single time — the scanner wakes hourly, so a card fires at the first wake-up
at or after its due mark. And the schedule assumes the app keeps running: the scanner only
advances while Omniscio is open, so a machine left closed for two days shifts everything later.
The preview reports the same rotation and the same on/off checks the scanner itself uses (queue.ts) rather than keeping its own copy of the rules, so it cannot drift into describing a schedule that never happens — a guard test fails the build if either side stops using them, and a second test runs both against the same settings and checks the preview's answer matches what the scanner actually did. (Before 2026-10-15 the on/off checks were written twice and the preview missed Tips & Guidance, so a default install was told a card was due when none would ever come.) It is also read-only: previewing never creates a card, never notifies you, and never marks a feature as shown.
Because the rotation is computed independently of the gates, the preview showed the full
schedule while nudges were switched off — which was the point: you could see exactly what
you would get before deciding to turn them on. With the feature retired there is nothing to
turn on: the preview answers only on an install where the dev reveal is active, and otherwise
reports flag-off. It never creates, notifies or marks anything, so it stays safe to call.
Relationship to the Workflow Coach
This is separate from the weekly AI Workflow Coach "feature-discovery" suggestions. The Workflow Coach is weekly, LLM-generated (costs money, needs an API key) and lives in the coach surface. Feature Discovery Nudges is the daily, deterministic, hand-curated (free) surface with a richer popup and one-click enable. They're complementary.
Related
Feature Suite Recommendations groups the same features into themed suites and is what the proactive recommenders consult; Empty-inbox discovery tip is the other one-time, teach-by-doing card Omniscio puts in the Inbox.
Last verified 2026-09-23