Omniscio

AI Manager

AI Manager is where you control the AI that runs at the end of each agent turn — Inbox Pilot routing and the Plain Speak overlay. It gives you the controls to pause an automation, edit its rule, check today's spend, or switch it off everywhere.

On this page

What it is

AI Manager is the oversight layer for Omniscio's turn-end AI pipeline. When an agent finishes a reply, two optional automations can act on it: the Inbox Pilot classifier, which sorts each turn, and the Plain Speak overlay, which rewrites a finished reply into a glance-readable card. AI Manager is the control surface that governs both.

It is the "who is watching the watcher" surface: because these automations spend money and change what you see, AI Manager gives you a single, obvious place to see their state, rein them in, and turn them off.

Where to find it

AI Manager is not a standalone panel and has no surface of its own. It is the shared AiManagerFeaturePanel component, rendered inside each feature's existing screen — the Inbox Pilot view (src/renderer/src/features/inbox-pilot/InboxPilotView.tsx) and the Plain Speak settings page (src/renderer/src/features/plain-speak/PlainSpeakSettings.tsx). Which controls appear depends on which feature you are looking at:

  • Inbox Pilot — the master on/off toggle, a Shadow mode (observe only) toggle, its editable routing rule (with rule-enabled toggle and a Preview action), the Cost & access card (daily cap and API-key allowance), and a recent-activity feed.
  • Plain Speak — the master on/off toggle, an Emojis on action labels toggle, Show Plain Speak first (default view), and Skip Plain Speak for phrases. Its rule card, cost card, and activity feed are all hidden: the inline card is free, so there is nothing to cap, and its prompt is locked to the built-in default.

Per-session toggles are not in this panel, and the two features differ. Each session's kebab (three-dot) menu carries an Inbox Pilot toggle and a per-session custom rule, but Plain Speak's per-session opt-out was removed on 2026-05-14 — Plain Speak now runs on every session whenever its master toggle is on.

How it behaves

Flipping a feature's master toggle stops that automation immediately; flipping it back resumes it. Editing a feature's rule changes how that feature behaves on the next turn. The daily-spend readout answers "how much has AI Manager spent today" for a given feature. Turning the Inbox Pilot master off is the broadest in-app lever: it resets every Inbox Pilot sub-setting and disables Inbox Pilot on every existing session in one action, rather than one session at a time. A separate global POST /ai-manager/pause-all route (CLI-token only) flips the top-level aiManager.paused flag for the whole system.

Private projects are never previewed. A Preview aimed at an AI Coaching or Journal session is refused, for both the routing rule and Plain Speak: those transcripts never leave the computer, so no preview model call is made and the modal shows the refusal instead of a result. Plain Speak's free inline card still applies in those sessions — the agent writes it, nothing is sent anywhere.

Because the router half is gated behind Inbox Pilot, some AI Manager routes only respond when Inbox Pilot is enabled; the overlay half is independent and survives Inbox Pilot being off.

Last verified 2026-10-06