Controls, permissions, privacy & general settings
Settings reference for what agents and external scripts are allowed to do, how data and activity are handled, and the productivity features that do not belong to a single integration.
On this page
- Agent permission level
- Require approval for AI session spawn
- Require approval: App settings
- Require approval: Session pause / snooze / archive
- Require approval: Tags
- Require approval: Project edits
- Require approval: Project / group delete
- Require approval: Away-Mode (Auto-Run) rules
- Require approval: Email Summarizer rules
- Require approval: Drip
- Require approval: Bug-Intake sources
- Require approval: AI Coaching
- Require approval: Recipe run
- Require approval: Nighty-Tidy run-now
- Require approval: Re-run from a message
- Require approval: Send SMS
- Require approval: Send a share link
- Require approval: Install / enable / disable / uninstall plugins
- Require approval: Start a screen recording
- Require approval: Screen-capture AI
- Require approval: Image generation and edits
- Require approval: Vault image AI analysis
- Require approval: Vault note summaries
- Require approval: Vault deletes
- Require approval: Tasks AI & agent runs
- Require approval: Meeting-note AI enhance
- Require approval: Send Gmail
- Require approval: Send Supermail / open assistant
- Require approval: Supermail panel settings and commands
- Require approval: Send Telegram messages
- Require approval: Admin actions
- Let agents bring Omniscio to the front
- Let agents switch your view directly
- Let agents restart the app
- Let agents spawn sessions with their own token
- Let agents write to Team Chat with their own token
- Let agents use Supermail with their own token
- Auto-approve config file writes
- Auto-approve plans
- Auto-fix master debt
- Redact pasted secrets
- Email diagnostics to the team
- Email crash reports
- Email bug reports & feedback
- Developer messages
- Measure how the window feels
- Performance & interaction tracing
- Track which UI controls you use
- Capitalize tasks automatically
- Automatically tidy away controls you never use
- Bookmarks CLI access
- Pre-compute codebase stats in the background
- Custom format block
- Log level
- Catch follow-ups when a session ends
- Bring in action items from Fathom
- Bring in action items from Fireflies
- Bring in action items from Meetings
- Whose action items from calls
- Tasks memory (learned planning context)
- Plan My Day
- AI arrange for the day’s plate
- Show day-load labels in Plan My Day
- Tasks side panel in Mission Control
- Show text on the collapsed Tasks panel
- Triage your Inbox list with an AI session
- Daily check-in
- Daily check-in hour
- Permanently remove deleted tasks after 30 days
- Tasks reminder alerts
- Due-soon alert (hours before deadline)
- Email me task reminders
- Text me task reminders
- Send task reminders to my phone
- Follow-up catching guidance
- Milestone nudges
- Email me a daily copy of the activity record
These settings control what agents and external scripts are allowed to do, how your data and activity are handled, and a range of productivity features that don't belong to a single integration.
Agent permission level
Controls how much a Claude agent can do on its own before Omniscio stops to ask you. Four levels are available: Read-only (agent can read files but must ask before writing or running commands), Guarded (auto-approves edits inside the project, asks for commands and outside-project writes — the recommended default), Autonomous (auto-approves edits and commands inside the project, asks only for writes outside), and Full trust (approves everything except the built-in protection for SSH keys, credentials, and system files, which always asks). Applies to Claude Code sessions, and to new Codex sessions, which get the closest Codex policy unless the Codex permission level overrides it — at Read-only, Codex runs in its own read-only sandbox and refuses every file edit. For a session running on another machine over SSH, "the project" means that session's folder on the remote machine (the one Claude reports when it starts), plus that machine's /tmp; until Omniscio knows that folder, every write from the session asks.
Require approval for AI session spawn
When ON, any AI-initiated session spawn — from a deep link or the CLI — appears as an inbox card you must approve before the session starts. When OFF (default), spawns happen immediately in the background and show in the sidebar without interrupting you. Turn ON if you want a human-in-the-loop checkpoint before AI orchestration can spin up new paid sessions.
Require approval: App settings
When ON (default), an agent or script trying to change any Omniscio setting via the CLI must wait for your inbox approval first. Governs both single-key changes and bulk settings-file imports. Turn OFF to let agents adjust settings without asking.
One exception. The Vault's knowledge-aggregation settings change through the Vault's own command-line route (PATCH /kms/aggregation/settings) and are applied immediately with no approval card, whatever this switch says. That route switches the paid daily or weekly digest on or off, sets the hours it runs, sets the per-run spending cap (anywhere from $0.01 to $10.00) and the AI deep-dive count. Starting a digest run on demand always asks; arming a recurring paid run and raising its cap does not. See Vault knowledge aggregation.
Require approval: Session pause / snooze / archive
When OFF (default), CLI requests to pause, unpause, snooze, or archive a session are applied immediately, so an agent can manage session lifecycle without asking. Turn ON to send those requests to your inbox for approval first — worth doing if you let agents you do not fully trust manage the fleet, since an ungated CLI once let an agent silently pause many sessions at once.
Require approval: Tags
When ON (default), CLI requests to create, rename, or delete library tags require your inbox approval. Turn OFF to let agents manage tags freely.
Require approval: Project edits
When ON (default), CLI requests to update a project's bug-intake settings or upload/delete project documents require your approval. Turn OFF to allow agents to edit project details without asking.
Require approval: Project / group delete
When ON (default), CLI requests to delete a project (soft-delete — recoverable) or remove a sidebar group (its projects become ungrouped) require your inbox approval. This is a destructive action. Turn OFF to let agents delete projects and groups without asking.
Require approval: Away-Mode (Auto-Run) rules
When ON (default), CLI requests to create, edit, or delete auto-run rules need your approval. Auto-run rules define what runs automatically while you are away, so changes to them have broad effects. Turn OFF to let agents manage these rules freely.
Require approval: Email Summarizer rules
When ON (default), CLI requests to create, update, delete, or toggle email-summarizer rules — including running the setup wizard or triggering a Gmail pass — require your approval. Turn OFF to let agents adjust summarizer rules without asking.
Require approval: Drip
When OFF (default), CLI requests to create, update, or delete Drip campaigns or their folder/book sources are applied immediately, so agents can manage Drip campaigns freely. Turn ON to require your approval first — worth doing if agents write to outbound campaigns you have not reviewed.
Require approval: Bug-Intake sources
When ON (default), CLI requests to enable a bug-intake source require your approval before going live. Enabling a source arms a recurring paid poll, so unexpected enables cost money. Turn OFF to let agents enable intake sources without asking.
Require approval: AI Coaching
When ON (default), CLI requests to create, edit, or delete coaching artifacts, or to start a coaching interview, require your approval. Starting an interview spawns a paid Claude session. Turn OFF to let agents manage coaching content and start interviews without asking.
Require approval: Recipe run
When ON (default), CLI requests to run a recipe (or arm a recurring schedule for one) require your approval, because running a recipe spawns a paid Claude session. Turn OFF to let agents trigger recipe runs without asking.
Require approval: Nighty-Tidy run-now
When ON (default), CLI requests to run a Nighty-Tidy audit immediately require your approval, because the run spawns a paid Claude session. Turn OFF to let agents trigger audits without asking.
Require approval: Re-run from a message
When ON (default), CLI requests to re-run a session from a specific message require your approval, because a re-run forks a new paid Claude session. Turn OFF to let agents trigger re-runs without asking.
Require approval: Send SMS
When ON (default), CLI requests to send or schedule a text message require your approval, because each SMS counts against your monthly limit. Turn OFF to let agents send texts without asking.
Require approval: Send a share link
When ON (default), CLI requests to deliver a share link (via email, SMS, or Slack) require your approval. SMS delivery counts against your monthly limit. Turn OFF to let agents send share links without asking.
Require approval: Install / enable / disable / uninstall plugins
When ON (default), CLI requests to install, enable, disable, or uninstall a plugin require your approval. Installing or enabling activates the plugin's UI, backend code, and any scheduled work; uninstalling deletes its files — both are significant changes. Turn OFF to let agents manage plugins without asking.
Require approval: Start a screen recording
When ON (default), a CLI request from an agent to start a screen recording — a screen take via POST /capture/recording/start — requires your approval first. Nothing is recorded until you approve the card; the click itself is the approval, so there is no second confirmation and no cost or delete to warn about. A non-agent caller using the CLI token is not gated here and starts immediately, as before. Turn OFF to let agents begin recordings without asking, or use the card's "always allow" caret to auto-approve this action from then on.
Require approval: Screen-capture AI
When ON (default), CLI requests to auto-title a snip or summarize a recording require your approval, because these call the paid Anthropic API. Turn OFF to let agents use screen-capture AI features without asking.
Require approval: Image generation and edits
When ON (default), an agent asking the CLI to generate an image, upscale one, or remove an image's background requires your approval, because every one of those spends the image credits on your account. Turn OFF to let agents do it without asking. The app's own Image Studio is unaffected — this switch is about the CLI.
Require approval: Vault image AI analysis
When ON (default), CLI requests to analyze Vault images using AI require your approval, because analysis calls the paid Anthropic vision API. Turn OFF to let agents run image analysis without asking.
Require approval: Vault note summaries
When ON (default), CLI requests to generate or bulk-seed Vault note summaries require your approval, because summary generation is a paid AI call that sends the note's text to OpenRouter, an outside AI service, which routes it to a hosted OpenAI model (openai/gpt-5.6-luna) — not to Anthropic (see Vault summaries). Turn OFF to let agents generate summaries without asking.
Require approval: Vault deletes
When ON (default), CLI requests to delete a Vault note, tag, or image require your approval. Recovery is limited, and only notes have any: a deleted note moves to the vault trash and can be restored from the Trash panel for at least 30 days (longer if your Data Retention setting is longer), after which the retention sweep permanently erases it — the trash is not included in backups either. A deleted image is removed for good, with no trash copy and no restore, and its AI title, description and OCR text go with it. Deleting a tag rewrites it out of every note that carries it. Turn OFF to let agents delete Vault content without asking, and remember that a silent delete by an agent may sit unnoticed until the trash window closes.
Require approval: Tasks AI & agent runs
When ON (default), CLI requests for paid Tasks AI actions — ranking, daily check-ins and task breakdowns — require your approval, because all of these call paid Claude. Launching a task agent is not on this switch at all: an agent launch always asks, whatever this is set to, because it starts a paid session. Starting an AI delegation run on a task is covered too: while this is on, an agent's request waits for your approval. One paid action is not covered: confirming the evening close-out, which starts the once-a-day "Learned about you" pass when Tasks memory is on; it does not ask first. Turn OFF to let agents run Tasks AI features without asking. This switch covers paid AI only: agents and scripts can delete, archive and move tasks — up to 100 in one request, or a whole project and its tasks — without asking, whatever it is set to. Deleted tasks are kept for 30 days and can be brought back from Recently deleted, opened from ⋯ More → View → Recently deleted in Tasks (details).
Require approval: Meeting-note AI enhance
When ON (default), CLI requests to AI-enhance a meeting note (Generate notes) require your approval, because enhancement calls the paid Anthropic API. Turn OFF to let agents enhance meeting notes without asking.
Require approval: Send Gmail
When ON (default), CLI requests to send, reply to, or forward an email via Gmail require your approval, because these send a real email to a real recipient. Turn OFF to let agents send Gmail without asking.
Require approval: Send Supermail / open assistant
When ON (default), CLI requests to send or reply to a Supermail email, or open a thread's AI assistant, require your approval. Sending emails reaches real recipients; opening the assistant spawns a paid Claude session. One toggle governs all three. Turn OFF to let agents do these without asking.
Require approval: Supermail panel settings and commands
When ON (default), CLI requests to write sensitive Supermail settings (AI-filter rule, filter mode, master on/off) or dispatch destructive commands (trash, sign-out, and similar) require your approval. This is a destructive category. Turn OFF to let agents manage Supermail settings directly.
Require approval: Send Telegram messages
When ON (default), CLI requests to send, edit, delete, or forward a Telegram message require your approval, because these are real actions a real person sees. Turn OFF to let agents interact with Telegram without asking.
Require approval: Admin actions
When ON (default), an agent or script asking Omniscio to do any of these must wait for your inbox approval first: claim the public support address, let (or stop) a project's cloud machine reading its repository, move sessions from one Overseer to another, start a session in an abandoned working folder, or delete one. Turn OFF to let them proceed without asking.
These six are the ones worth reading before you turn anything off. One of them spends money (starting a session in an abandoned folder), and one destroys work that may exist nowhere else (deleting that folder). The other four change what a machine can reach: enabling cloud repo access writes a deploy key onto a repository you own, disabling it revokes one, moving sessions re-homes work that is already running, and claiming the support address re-points a door anyone can email to start a session.
You can also allow just one of them. Every approval card for these carries an "always allow" option, so approving one command once and asking to stop being asked changes that command alone — the others keep asking.
Let agents bring Omniscio to the front
When OFF (default), an agent's request to focus or pop Omniscio to the front while you're working in another app is ignored — instead you get a dismissible inbox note naming the session that tried. When ON, agents may pull the app to the front from behind. Your own shortcuts, notification clicks, and phone taps are never affected.
Let agents switch your view directly
When OFF (default), an agent's request to jump you to a hub, the inbox, a settings pane, a note, a mind map, or a Gmail thread shows a click-to-go toast rather than moving you immediately — so Omniscio never takes over your screen without permission. When ON, agents may switch your view directly (useful for guided tours and demos). If you're in another app entirely, the request still can't move you — you get an inbox note instead.
Let agents restart the app
When OFF (default), the CLI restart route is disabled entirely — no agent or external script can quit and relaunch Omniscio, even one holding the global CLI token. Your own "Restart Omniscio" button in the app is unaffected. Turn ON only if you need a headless script to restart the app.
Allowing just one session instead. You don't have to open this up for everyone. In a run-from-source (dev) build, open any session's ⋯ menu and tick Allow restarting Omniscio — that one session may then restart the app when it needs to (for example, to load code that just landed), and every other session stays blocked. A session can also ask: you get an Approve / Decline card in the inbox with its reason. The permission lasts until you untick it; an Overseer keeps it through its daily refresh, and a session that hands off passes it to its successor. No agent can give itself this permission.
Let agents spawn sessions with their own token
When OFF (default), a session-spawn request is rejected if it presents only an agent's scoped token (the Omniscio-injected $AMC_CLI_TOKEN). This stops a compromised MCP server or npm dependency that reads an agent's environment from spending your money by spawning paid sessions. Agents still spawn normally with the global CLI token. Turn ON to allow agents to spawn sessions directly with their own scoped token.
Let agents write to Team Chat with their own token
When OFF (default), Team Chat writes over the CLI (send/DM, react, edit, delete, pin, schedule, invite) are rejected if they use only an agent's scoped token — preventing a compromised dependency from messaging your teammates as you. When ON, agents may write to Team Chat with their own token; every message is still labeled as coming from your agent. Reading Team Chat is unaffected.
Let agents use Supermail with their own token
When OFF (default), Supermail's mailbox over the CLI — reading, searching, archiving, labeling, snoozing, and requesting a send or reply — refuses an agent's own scoped token, including a cloud session's, so a compromised add-on that reads an agent's environment can't read your mail. Turn ON to let your agents, including cloud sessions, work in your inbox. Sending or replying still follows your CLI email-approval setting, and mail rules, the AI filter, blocked domains, snippets, splits and the paid Supermail assistant stay yours alone. An agent that reads an email can be steered by what it says, so only turn this on if you want agents in your mail.
Auto-approve config file writes
When ON, Omniscio automatically approves when Claude wants to modify its own configuration files (.claude/ directories and CLAUDE.md). When OFF, you are asked to approve each config-file change before it is written. Useful to turn ON in a heavily automated workflow; leave OFF if you want to review every instruction-file edit.
Auto-approve plans
When ON, Omniscio automatically approves the plan Claude proposes before writing code, skipping the approval step entirely. When OFF (default behavior without this toggle), you are prompted to approve or reject each plan first. Saves a click in high-volume, trusted workflows. If the session's agent has already stopped by the time the approval is sent (cloud sessions stop after every turn), Omniscio restarts the session to deliver it; if it cannot, the plan lands in your inbox to approve by hand rather than the session sitting stuck.
Auto-fix master debt
When ON, Omniscio automatically spawns fix sessions for failures already present on the base branch you are working from, rather than leaving them for your branch to carry. When OFF (default), you receive an inbox card asking you to approve each fix run first. Requires the Master-Debt Auto-Fixer feature to be enabled separately. Because each fix run is a paid session, the OFF default lets you control the spend.
Redact pasted secrets
When ON, API keys and secrets you paste into a chat message are replaced with a placeholder before the message is saved to disk, so your keys are not stored in plain text in the local database. The agent still receives the real key in the outgoing request so it can act on it. Turn OFF only if the redaction interferes with a specific workflow.
Email diagnostics to the team
Master switch for Omniscio's outbound diagnostic emails from this installation. When ON (default), a weekly usage digest is sent, plus individual crash reports (unless you turn off the crash email below). When OFF, both are silenced. On-disk crash logs, Sentry error triage, and all in-app diagnostics continue regardless.
Email crash reports
When ON (default, and requires diagnostic emails to be on above), Omniscio emails the team on each crash as it happens. When OFF, only the weekly digest is sent — crashes still auto-triage inside Omniscio via Sentry, and the local crash log continues unchanged.
Email bug reports & feedback
When ON (default), clicking the bug-report icon, sending feedback, or reporting a missed question widget sends your report to the Omniscio team through two channels: email (Resend) and the bug database (Firestore).
When OFF, none of them fires — the report is written to the local disk fallback
(<userData>/bug-reports/…json, the same outbox the retry drain reads — its 20 most recent reports
are kept) and stays on your computer; the retry drain sends nothing while the switch is off, and the
kept reports go out once you turn it back on.
Changed by the privacy audit. This toggle used to gate the email leg ONLY, while a submission actually fanned out to both destinations — so the one control a user was offered silenced one of two, and the copy said so out loud. Firestore's only gate was
bugReportTransport, which was declared in the settings types but wired into no settings schema, so neither the UI nor the CLI could change it. The switch now means what it says.
Developer messages
When ON (default), you receive announcements, surveys, and read-receipt events from the Omniscio team inside the app. When OFF, all of these are silenced from this installation.
Measure how the window feels
When ON (default), Omniscio measures how long your clicks and keypresses take and how long the window is blocked by slow frames, so it can tell when it is slowing you down rather than guessing. Only slow events are recorded — a fast click leaves no trace — and the measurement is passive: your own machine's browser does the timing. This is what feeds the interaction-latency, long-frame and blocked-seconds readings on the performance readout and the capacity model. Turning it OFF stops the measurement and those three readings go back to "not measured" rather than reporting a zero. Nothing is sent anywhere — all data stays on your computer.
Performance & interaction tracing
When ON, Omniscio also attributes each slow frame to the code that caused it and keeps a click-by-click trace in the local troubleshooting log files (renderer diagnostics and interaction.log). Off by default, because this half runs code inside every animation frame and every click you make — it is the expensive part, and it only adds the per-frame attribution and the bug-report trace, never a reading. Switch it on before you reproduce a freeze if you want that attribution to exist. Your crash reports, error logs, the freeze recorder and the "Measure how the window feels" measurements above stay on regardless. Nothing is sent anywhere — all data stays on your computer.
Track which UI controls you use
When ON (default), Omniscio records which buttons, menus, and sidebar integrations you actually use, powering the "UI Usage" panel where you can see what is used versus never used and decide what to hide or remove. The report stays on this device. Anonymous click counts also ride the weekly diagnostics digest when diagnostic emails are on. Turn OFF to stop all usage recording.
Capitalize tasks automatically
When ON, the first word of a task and the start of each sentence are auto-capitalized as you type — skipping URLs, file paths, and abbreviations. Matches the behavior of the Vault notes app. On by default; turn it OFF if you would rather capitalize manually.
Automatically tidy away controls you never use
When ON, Omniscio moves sidebar integrations and toolbar icons you have not used in 30 or more days into a quiet "Unused" group or the "…" overflow to reduce clutter, and notifies you in the inbox. Nothing is deleted — you can restore anything from the list at any time. Off by default; turn ON if you want the sidebar to stay lean on its own.
Bookmarks CLI access
When ON (default), external agents can list, create, update, delete, reorder, and launch bookmarks over the local CLI control server. Turn OFF to disable all bookmark CLI access from this installation at once without touching individual bookmark settings.
Pre-compute codebase stats in the background
When ON (default), Omniscio quietly analyzes and caches each hub's codebase stats (size, languages, git activity, code health) in the background so the "Codebase Stats" view opens instantly. The analysis is heavily throttled — one hub at a time, only changed content, and it backs off when the machine is busy — so it never slows things down. Dependency security scans still run only on demand when you open a hub and refresh.
Custom format block
When ON, Omniscio silently adds a short, editable formatting standard to the first message of each new session, alongside the question-widget hint, so the agent shapes its responses to be easy to skim. You can edit the exact text in the field below. Turn OFF to stop the hint from reaching the model entirely.
Log level
Controls how much detail is written to the local log file. Options (each includes all levels below it): Error — critical failures only; Warn — errors and warnings; Info — general operational events (recommended for everyday use); Verbose — detailed operational events, useful when troubleshooting; Debug — everything, including low-level internals. Default is info. Switch to verbose when investigating an issue, then switch back to avoid log growth.
Catch follow-ups when a session ends
Off by default — this is opt-in, because it sends your session text to an outside AI service. When ON, after a session ends or is archived the end of its transcript (the last ~12,000 characters, so on a long session only the final few exchanges; earlier follow-ups are not seen) is sent first to your own Claude account (Anthropic), and only when that account cannot answer is it sent to OpenRouter, an outside AI service that routes the call to a hosted model (or to a saved key or credits you set up), to look for follow-up tasks and stage them in the Caught view for you to approve or dismiss. Each scan also sends the titles of up to 40 of your open tasks from every project (each cut to about 120 characters), so the AI can spot ones you've finished, plus a short note on which tasks you keep carrying over. When OFF, nothing is sent — no transcript is read and no AI spend is incurred. It also gates action items from your calls: with it off, nothing is added to Caught from Fathom, Fireflies or Meetings either, and the per-source switches below are grayed out. Meeting items come from the notes you already made, so they send nothing extra to an outside service. Requires the Tasks feature to be active. Each person on a shared computer has their own choice.
Bring in action items from Fathom
Off by default. When ON, Omniscio checks your Fathom meetings about every 15 minutes (meetings from the last 3 days, at most 20 per check) and stages the action items that are yours in the Caught view, each labeled "from Fathom meeting" with the meeting's name and a link that opens the recording at the moment it was said. Only your own items arrive unless you change Whose action items from calls. A meeting is read once, and nothing is read while this switch, Caught or Tasks is off. It does nothing until Fathom is connected with your key (Settings → My API Keys), and it is grayed out while Caught is off. Fathom is read with your existing key and nothing new is sent anywhere. Checking happens on this computer only, and what it catches stays on this computer. Each person on a shared computer has their own choice. Find it under Settings → Tasks (the Bring action items from card) or the Sources button in Caught.
Bring in action items from Fireflies
Off by default. When ON, Omniscio checks your Fireflies meetings about every 15 minutes (the last 3 days, at most 20 per check) and stages the action items that are yours in the Caught view, each labeled "from Fireflies meeting" with the meeting's name and a link to its transcript. An item counts as yours when Fireflies names you in full; an item with no name, or only a first name, is kept so nothing is lost. A meeting is read once, and nothing is read while this switch, Caught or Tasks is off. It does nothing until Fireflies is connected with your key (Settings → My API Keys), and it is grayed out while Caught is off. Checking happens on this computer only, and each person on a shared computer has their own choice. Find it under Settings → Tasks (the Bring action items from card) or the Sources button in Caught.
Bring in action items from Meetings
On by default, so people who already received meeting action items in Caught keep receiving them. When ON (and Caught is on), the action items in a meeting's AI notes arrive in the Caught view as suggestions marked "from meeting" with the meeting's name. Turning it on from Settings or the Sources button in Caught also turns on automatic AI notes, so every recorded meeting makes its notes, and its action items, without a click. That costs one small AI call per meeting, shown under Meetings in your spending, and it uses the same AI Meetings already uses for notes. Turning it off stops meeting items from arriving but does not turn automatic notes off. It only appears while Meetings is turned on, it is grayed out while Caught is off, and each person on a shared computer has their own choice.
Whose action items from calls
Chooses which action items from Fathom, Fireflies and Meetings arrive in Caught. Only mine (the default) keeps items assigned to you, items assigned to no one, and items whose owner is unclear, and leaves out items clearly assigned to someone else. Everyone's keeps every item. Items already marked done never arrive. Each person on a shared computer has their own choice. It sits at the bottom of the Bring action items from card under Settings → Tasks and of the Sources panel in Caught.
Tasks memory (learned planning context)
When ON (off by default for a new install), Omniscio journals what you planned versus finished each day so the AI can learn your real patterns — typical daily capacity, chronic carry-overs, and tasks you keep snoozing. Reading the stored journal is free. A daily distill pass runs at day-close to update your "Learned about you" notes; this is a small paid AI call. Those notes are kept in About You, stored only on this computer, and when Global Memory is on they also help your other sessions. The arrange and check-in features also make AI calls when they actively use that context. When OFF, nothing new is journaled (no arrange-time plate snapshot, no close-out outcome) and the distill pass is skipped; confirming the close-out still marks the day reviewed. An install that had saved its settings before this switch existed, and never touched it since, comes up with it ON and keeps journaling and distilling until you turn it off under Settings → Tasks → Memory (learned planning context). See what it has learned under Learn about your habits on Today. Each person on a shared computer has their own choice.
Plan My Day
When ON (default), Tasks offers Plan My Day: every Plan my day button starts a planning session that reads your tasks and your Today plate and plans the day with you, on your selected AI model. That session is billed to your own AI account. When OFF, every Plan My Day AI call is blocked, including that session, and the Show day-load labels in Plan My Day row is grayed out. Opening Plan My Day from Tasks while it is off asks whether to turn it on. This applies to everyone on this computer.
AI arrange for the day’s plate
When ON (default), arranging your Today plate can use the stronger AI pass to reorder your day. Each arrange is one paid AI call (on your Claude account first, then your saved keys or credits); there is no Tasks spending limit, so turn this off if you do not want arrange to spend. When OFF, Arrange stays free: it still sorts by your pins and due dates but never calls the AI, and the line under the Today toolbar says "AI is off, so this uses your own task list" with a Turn AI back on button. This applies to everyone on this computer.
Show day-load labels in Plan My Day
When ON (default), Plan My Day shows a small label above its AI message that sums up how full the day looks: Light day, Normal day, Heavy day or Overloaded day. Tasks grades it by comparing the tasks you have open — not only the ones due today — with how many you usually finish in a day, and leaves it off when it cannot judge. Heavy days show the label in a warning color and overloaded days in a danger color, so a packed plan stands out before you accept it. When OFF, only the label is hidden: Plan My Day plans your day exactly the same way, so this switch does not change what is spent. This applies to everyone on this computer.
Tasks side panel in Mission Control
When ON, Mission Control shows a My Tasks panel along its right edge: your Tasks list, with the number of open tasks in its header, so you can work through your own tasks beside the board. It only appears while Tasks itself is also on. Drag the panel's left edge to make it wider or narrower, and use the sun or moon button in its header to switch it between light and dark. The arrow button at the foot of the panel collapses it to a slim My Tasks strip, and clicking the strip opens it again; the panel remembers whether you left it collapsed. The strip shows the sky-blue Tasks icon under its arrow (the same icon the open panel's header and the sidebar's Tasks row use), the words My Tasks centered below it, and your open-task count at the bottom. Collapsing only tucks the panel away. To remove it from Mission Control, turn this setting OFF. Choosing Keep Tasks as it is, show it beside Mission Control in the Your tasks step of the move from Tasks to Mission Control turns this setting ON for you. Default is OFF.
Show text on the collapsed Tasks panel
When ON (default), the slim My Tasks strip that Mission Control's Tasks side panel collapses to shows the words My Tasks, written sideways and centered. When OFF, the strip shows only its arrow, the Tasks icon and the open-task count; hovering it still names it, screen readers still announce it with the count, and clicking it still opens the panel. It is in the Tasks settings and appears whenever Tasks is on; it only changes anything while the Tasks side panel in Mission Control is on and collapsed.
Triage your Inbox list with an AI session
When ON (default), you can start a project-manager AI session from the Caught lane's Triage this for me button (set up sorting rules, or group your caught items), and from the Inbox triage card that files your unsorted Inbox tasks. Each only runs when you start it. When OFF, the Caught lane hides its Triage button and the app refuses both triage sessions in the main process, so neither can be started from any surface, and the Inbox triage card is hidden; your caught items and Inbox tasks stay where they are for you to sort by hand. The card is also hidden while Tasks itself is off. This applies to everyone on this computer.
Daily check-in
When ON, Tasks starts a short AI check-in conversation once a day on its own: it reviews what was recently finished, what is due and what is slipping, and proposes today's focus for you to accept. The focus it proposes is at most 3 tasks, even when your planning profile's preferred daily count is higher; Plan my day plans a longer list. It starts at the hour set in Daily check-in hour (below) and runs at most once a day — if you already opened today's check-in yourself, nothing extra runs. The app has to be open for it to start. Default is OFF, because it spends on the AI, so you opt in. You can always open a check-in by hand from Tasks whether this is on or off. Each person on a shared computer has their own choice.
Daily check-in hour
The hour of the day (0–23, in your local time) at or after which the automatic daily check-in starts. Default is 8, so the check-in starts from 8 a.m. onward. Tasks checks about every 15 minutes, so it can start up to a quarter of an hour after the hour you pick, or later if the app was closed at that time. It only matters when Daily check-in (above) is turned on. Each person on a shared computer has their own choice.
Permanently remove deleted tasks after 30 days
When ON (default), a task you delete stays recoverable for 30 days, then is permanently removed — together with its lines in the day journal — so a deleted task does not live on in the app or in your backups forever. Restoring a task within the 30 days clears its clock, and it is never removed. The cleanup runs once a day on its own. When OFF, deleted tasks, the projects you deleted, and the files attached to them, are kept forever instead; the rest of the same daily cleanup is unaffected, so deleted check-ins, the conversation and plan of older check-ins, removed notes and restarted interviews are still cleared after a fixed 30 days (the Data Retention setting does not change this), and reset or dismissed learnings are cleared on their own fixed 30-day window. Only Caught suggestions follow your Data Retention setting. This applies to everyone on this computer.
Tasks reminder alerts
When ON (the default), Tasks raises its reminder cards in your Inbox: a task is due soon, a task is due, a snoozed task is back, and time to follow up on a task you are waiting on. Turn it OFF to stop all four at once, along with the phone, email and text reminders that come with them. Snoozed tasks still come back into your list on time, just without a card, and follow-up cards for finished tasks are still cleared. Turn it off from the Turn off these alerts button on any of those cards, and back on in Settings → Notifications → Alert types. When you turn it back on, any task that is still overdue raises its card again, a few at a time. This applies to everyone on this computer.
Due-soon alert (hours before deadline)
How many hours before a task's deadline Tasks sends an early heads-up to your Inbox. Default is 24 hours; you can set anything from 0 to 168 (one week). Set it to 0 to turn due-soon alerts off entirely — the alert when a deadline actually arrives still comes through. Each person on a shared computer has their own choice.
Email me task reminders
When ON, Tasks also emails you its reminders: a task coming due, a task that is due, a snoozed task coming back, and a follow-up nudge. They go to your agent email address, so you need to be signed in and have an agent email address set. Quiet hours apply, and repeat reminders for the same task are held back. Default is OFF. Without an agent email address the switch stays off and says so. The reminder still lands in your Inbox either way. This applies to everyone on this computer.
Text me task reminders
When ON, Tasks also texts you its reminders (due, due-soon, snooze-return and follow-up) to your alert phone number through Twilio, so Twilio has to be set up first. Until it is, the switch stays off and says so. Quiet hours apply, and repeat reminders for the same task are held back. Default is OFF. The reminder still lands in your Inbox either way. This applies to everyone on this computer.
Send task reminders to my phone
When ON, Tasks buzzes your paired phone with its reminders (due, due-soon, snooze-return and follow-up), and with two Tasks cards: "An AI draft is ready" (tapping it in the web app opens that task's Review draft panel) and "Task agents keep failing to start". It sends to a paired phone and to any browser or home-screen web app that turned on notifications through Web Access; with neither, nothing happens. Quiet hours apply, notifications you have silenced stay silent, and a repeat reminder for the same task is not sent again. With content-free phone notifications on (the default), the phone app shows generic text ("A task needs you") instead of the task itself, but a browser notification still shows the task's own text. Default is ON. The reminder still lands in your Inbox either way. This applies to everyone on this computer.
Follow-up catching guidance
Optional free-text guidance that steers which follow-up tasks the AI "catches" from your finished sessions. Describe, in your own words, what makes a follow-up worth keeping — for example "only things I still owe someone" or "skip anything about code." The AI uses it to decide which suggestions to keep; the format it returns is fixed, so rewording this can never break catching. Leave it empty for the standard behavior. It only matters when follow-up catching (above) is turned on. Accepted values: any text up to 2,000 characters. Each person on a shared computer has their own choice.
Milestone nudges
When ON (default), Omniscio shows a brief toast at natural progression points — your 5th task, your first overdue task, your 10th task, and your first project — each suggesting a next step you might not have discovered yet. Each toast has a primary action button that jumps you straight to the relevant feature and a "Don't show these" secondary action that turns this setting off. To get the tips back, turn Milestone nudges on again in the Tasks Caught settings (search Settings for "Milestone nudges"). Once a milestone fires it never fires again, even if you turn the setting back on later. The system seeds silently on first run so milestones you already passed before the feature shipped never announce retroactively. This applies to everyone on this computer.
Email me a daily copy of the activity record
This switch controls only the EMAIL. The activity record itself is always kept on this computer, on or off, and you can read it any time under Settings → CLI Control → Recent CLI Activity.
This switch is OFF by default: a brand-new install keeps its activity record on this computer and emails nothing until you turn the switch on here. When ON, Omniscio emails you the previous day's activity record once a day: what agents and other programs did here through its own control server — every session started, every inbox alert raised, every protected action taken. Only the actions are copied in full; the constant background reads are summarized as a count per kind of request, so the busy polling an agent does never crowds out the things worth remembering. A busy day arrives as up to three emails (part 1 of 3, and so on), and after Omniscio has been closed it catches up on the missed days, one per hour. Each copy is chained to the one before it with fingerprints, so an edited or missing line in a copy is visible, and the copies still exist somewhere else if the record on this computer is edited or wiped. Email addresses, keys and tokens are removed before anything is sent. A day that cannot be sent for 24 hours raises one inbox card, and a day that was changed after it was copied raises another. Turning this off stops only the emails and, within the hour, clears the "not leaving this computer" warning card; it does not delete the record on this computer or anything already sent. A "changed after it was copied" card stays until you put it away. This switch is deliberately not changeable from a phone or from the CLI, so a stolen device or a wandering agent cannot switch the emailed copy off.