Omniscio

More settings (B–G)

Plain-language reference for every Omniscio setting whose key begins with the letters B through G, presented in ordinary wording rather than technical detail.

On this page

Plain-language reference for every Omniscio setting whose key begins with the letters B through G.


Include screen recordings in backup mirror

Controls whether the Backup Mirror bundles your screen recording files when it syncs. Default is off. Screenshots are always included regardless of this setting; only the larger video recording files are gated here. A single recording can be 1–3 GB, and the mirror assembles the bundle in memory, so bundling recordings by default would risk running out of memory and bloating every sync. Turn it on only if you want irreplaceable recordings backed up and your system has enough RAM for the operation. Toggle; accepted values: true | false.


Enable Beautify Site skill

Adds the Beautify Site skill to your Claude Code skills. When you ask an agent to make a page or website look better, this skill walks it through one steady process: pick a single design direction, fit the page's parts to it, add motion, then check the finished page and remove the tell-tale signs of AI-generated design. Turning it on installs the skill the next time you launch Omniscio, and turning it off removes it again. You find it in Settings > Features.

Default: off.


Bug intake append prompt

Free-form operator text that is appended to the agent investigation prompt for every bug report the intake system receives — whether it arrives by email, Sentry, or the in-app form. Stacks before any per-project append prompt. Use it to inject standing instructions ("Always check the error logs first" or "CC the back-end team in your report"). Empty by default, meaning no extra instructions are injected. Text field; accepted values: string (≤10,000 characters).


Bug intake

Shows the Bug Intake entry in the sidebar — a feature that collects bug reports from configured Sentry sources, deduplicates them, and triages each one by spawning an agent session. When off, the Sentry poller is paused and all intake request handlers refuse connections. Off by default. This switch is now the Bug Intake Install in Marketplace → Inbuilt rather than a Settings row; accepted values: true | false.


Pace heavy worktree-drive work

Keeps Omniscio's own big file jobs from freezing your computer. These are the jobs that clear out retired worktrees (the separate working copies Omniscio makes for coding agents) and build the shared dependency folders on the Dev Drives. With this on, they run one at a time at a steady rate, wait out the first minutes after a restart, and pause while Windows is busy. Without it, the whole computer could lock up for 10 to 20 seconds, which happened most often right after a restart. Turn it off only if you want the older, unpaced behavior. It mostly matters if you use Omniscio's worktrees for coding work. You find it in Settings > Performance.

Default: on.


Today's Agenda card

Posts one "Today's Agenda" card to your inbox each morning, listing today's calendar events — each one a link that opens or edits it in Google Calendar. On by default, and it only runs while the Google Calendar integration is also on. You get at most one card per day, so a restart or a re-sync never gives you a second copy. Turn it off to stop the daily card without affecting the "starting soon" reminders. Toggle; accepted values: true | false.

Week starts on

Which day the calendar grid starts on. Monday is the default because that is what the calendar has always done — a Sunday default would have quietly shifted everyone's grid on upgrade. Choose Sunday if that matches how you read a week. Display only; it does not change any event's time or any reminder. Select; accepted values: "monday" | "sunday".

Calendar notification delivery

Chooses where calendar event reminders appear. inbox (default) creates a dismissible alert card in your inbox. toast shows a transient pop-up that disappears on its own. both does both at once. Change this in Settings → Notifications if you find one channel too easy to miss or too distracting. Select; accepted values: "inbox" | "toast" | "both".


Calendar reminder lead time

How many minutes before an event starts Omniscio fires the "starting soon" reminder. Default is 15 minutes. Set to 0 to be notified exactly at start time; the maximum is 120 minutes (two hours). Only fires while the Google Calendar integration and background sync are enabled. Number; accepted values: integer 0–120.


Chat depth style

Sets the visual depth style of the chat message list. flat renders messages with no card lift. soft adds a light surface separation between messages. glass uses a translucent frosted treatment. This is a display-only preference with no effect on performance or functionality. Select; accepted values: "flat" | "soft" | "glass".


Utility AI model

The single model that all of Omniscio's background AI helpers use — session title generation, reply suggestions, summaries, digests, and similar lightweight tasks. Automatic (the default) uses DeepSeek V4 Flash on your own DeepSeek key when one is configured (cheaper, billed to you), and falls back to the built-in model otherwise. Set a specific model ID here to override that choice everywhere. Advanced setting; found in Settings → Sessions. Text; accepted values: string (≤100).


Offer to move Claude Code to Anthropic's installer

If Claude Code on this computer was installed with winget or npm, this shows a one-click button on the Claude Code row that moves you to Anthropic's own copy. It installs the new copy first, keeps the old one until the new one is proven to run, and removes the old one afterwards. If a session is still using the old copy, the removal waits until the next time Omniscio quits. Turn this off if you do not want the button offered. You find it in Settings > Connected Tools.

Default: on.


Keep running when the window is closed

When enabled, closing the Omniscio window keeps the app and all your sessions alive in the system tray rather than quitting. Reopen the window from the tray icon. Default is off — closing the window quits the app. If you frequently run long background sessions, turning this on ensures they keep running while you work in other apps. Toggle; accepted values: true | false.


Keep cloud boxes running when you quit

Whether the cloud machines Omniscio started stay up when you quit the app. On (the default) leaves a box that is still inside its keep-warm window running after you quit, so a session you come straight back to does not pay for a fresh start — an unattended box then stops on its own idle timer rather than this app stopping it. Off restores the earlier behaviour exactly: quitting flushes every warm box, so nothing keeps billing after you quit. Read at the moment of the quit, so a change applies to the next quit with no restart. Toggle; accepted values: true | false.


Also review a cloud session's work when it comes home

This applies to cloud sessions, which are still in development. When a cloud session's finished work is brought back to this computer, this runs the AI Code Review reviewers you set up in the Dev Pipeline panel against the returned branch and records what they found. The findings are delivered to you and kept with the branch, and a review does not hold the branch back from landing. It is off because a cloud session already asks for a review on the cloud machine while it works, so returned work is normally reviewed there. Turn this on only if you want an extra review on top of that. You find it in Settings > CLI Control.

Default: off.


Let a cloud session read other sessions' transcripts

This applies to cloud sessions, which are still in development. A cloud session is an agent running on a cloud machine. When this is on, that agent can read the conversation of a session other than its own, and every one of those reads is recorded. Passwords and keys found in what it reads are blanked out first. When it is off, a cloud session can read only its own conversation. Your other sessions may hold anything you or anyone else pasted into them, so turn this off if you want a cloud session kept to its own conversation. Only you can change this switch, and an agent can never turn it back on. A change takes effect on the next read, and nothing already read is undone. You find it in Settings > CLI Control.

Default: on.


AI coaching

Enables the AI Coaching engine, which provides contextual tips and nudges based on how you work. Off by default. Turn it on in Settings → AI Coach if you want coaching input. Related settings let you choose which coaching categories are active and whether coaching artifacts mirror into your KMS vault. Toggle; accepted values: true | false.


Codex approval mode

Sets how much autonomy Codex sessions have when executing commands. inherit (default, shown as "Follow Agent Permission Level") follows the Agent Permission Level: Read-only runs Codex in its own read-only sandbox, which refuses every file edit and still asks before each command; Guarded acts like ask (full machine access, asking every time); Autonomous acts like ask-on-request plus internet access; Full trust acts like auto. ask grants full machine access but asks before every command. ask-on-request allows workspace-level writes freely but asks before reaching outside the workspace — including any command that needs the internet, because Codex's workspace sandbox has no network. auto grants full access and never asks. Pick a value other than inherit only to run Codex at a different level from Claude. Changes apply to newly started Codex sessions. Select; accepted values: "inherit" | "ask" | "ask-on-request" | "auto".


Codex virtual project

Shows a dedicated Codex virtual project in the sidebar so you can run OpenAI Codex sessions alongside Claude sessions. Off by default; requires a Codex API key configured in Settings → Accounts. Toggle; accepted values: true | false.


Concurrent agent shell commands

How many agent Bash and PowerShell calls are allowed through at once while "Queue concurrent agent shell commands" is on. Extra calls queue for up to 20 seconds and then run anyway. Choose a whole number from 1 to 512. This box only appears while the queue is switched on, and changes apply while the app is running. You find it in Settings > Performance, among the advanced options.

Default: 24.


Queue concurrent agent shell commands

A trial control (marked beta) that limits how many shell commands your agents run at the same time. It covers Bash and PowerShell calls from Claude sessions that Omniscio started. When the limit is reached, later calls wait for a free slot. A waiting call runs anyway after 20 seconds, or right away if the hook service is unavailable, so nothing is blocked for good. Your own terminals and other agent tools are not affected. Changes apply while the app is running. The limit itself is set by "Concurrent agent shell commands", which appears once this is on. You find it in Settings > Performance, among the advanced options.

Default: off.


Ask before archiving ends a running audit

When a session is still running a Nighty Tidy audit, its whole result lives in that session's work, so archiving it ends the audit and the findings are gone — two paid runs were lost this way. With this on (the default), closing such a session asks once first: the sidebar middle-click, the header X, the overflow menu, the keyboard shortcut, a swipe on your phone, a spoken command, and a bulk archive all show the one question. Every other session archives exactly as it does today, with no question and nothing slower. Turn it off and every session archives straight away, audits included. Archive anyway, Undo, and the rest of archiving are unchanged. You find it in Settings → Sessions → Behavior, beside "Warn before archiving loses work".

Default: on.


Content-free mobile push

When enabled (the default), mobile push notifications strip the session name and any other content before the notification is sent to the push provider — the provider sees only ciphertext and transport metadata, never the actual session text. Turn it off only if your push provider already encrypts everything end-to-end and you want richer notification previews. Toggle; accepted values: true | false.


Warn at 75% context full

Posts a note in the chat once a conversation has used about 75% of the model's context window, which is the amount of text the model can keep in mind at one time. The note reads "Context is N% full" with the token counts. It is only a heads-up, and nothing is broken. It appears at most once as a conversation fills, it is skipped if the 90% note has already appeared, and it is held back while the agent is parked waiting. Some models with very large windows do not show it. The note also opens a small panel with the same switch. You find it in Settings > Sessions.

Default: off.


Warn at 90% context full

Posts a note in the chat once a conversation is about 90% full. That is about the point where Claude may start summarizing and dropping older messages to keep going, so later replies can lose earlier detail. The note reads "Context is N% full" with the token counts and is only a heads-up. It appears at most once as a conversation fills, and it is held back while the agent is parked waiting. Some models with very large windows do not show it. You find it in Settings > Sessions.

Default: off.


Catch-Up Card

RETIRED — paused 2026-04-30, retired for good on 2026-10-06. The Catch-Up Card is no longer a live feature. On 2026-04-30 the Settings section this switch lived in was deleted, which left the feature switched off for everyone with no control to turn it back on; the owner then retired it outright on 2026-10-06. The card does not render above a session, its Settings controls are on no screen, and its search entries were removed on 2026-10-29, so Settings search no longer offers it either. The key and the five sections below it still exist in the schema and describe what the feature did; nothing in the interface reaches them. See catch-up-card.md.

Pins a short catch-up card at the top of any session that is sitting in "Needs You," giving you a 4-line summary of what the agent did since you last looked. Off by default — it costs an AI call each time the card is generated. There is no way to turn it on. Toggle; accepted values: true | false.


Catch-Up Card model

Which model generates the catch-up card summary. Defaults to Sonnet 4.5 when unset. Set a specific model ID to use a different (for example, cheaper or faster) model for these lightweight summaries. Only applies when the Catch-Up Card feature is enabled. Text; accepted values: string (≤200).


Catch-Up Card "originally" length cap

Maximum character count of the "what I originally did" section in the catch-up card prompt. Keeps the summarization input from ballooning on very long sessions. Default is 1,000 characters; range is 50–10,000. Increase if the truncated context causes the summary to miss important earlier context. Number; accepted values: integer 50–10,000.


Catch-Up Card "originally" prompt template

The editable prompt template used to generate the "what I originally did" part of the catch-up card. Leaving it unset uses the built-in default template. Edit only if you want to change the phrasing or structure of that section of the summary. Text; accepted values: string (≤50,000).


Catch-Up Card recent turns visible

How many of the most recent conversation turns are shown in full below the catch-up card summary, giving you immediate context alongside the generated overview. Default is determined by the built-in constant; range is 1–5 turns. Increase if you need more recent history visible at a glance. Number; accepted values: integer 1–5.


Show agent summary in Catch-Up Card

When on, the catch-up card also includes the agent's own self-reported summary (if the session produced one), in addition to the generated overview. Off by default. Toggle; accepted values: true | false.


AI Council evaluator model

The model that plays the "evaluator" role in an AI Council run — the blind critic agents that score each builder's attempt against the goal. Unset uses the Council's built-in default. Override here to use a specific model for evaluation across all Council runs. Text; accepted values: string (≤120).


AI Council synthesizer model

The model that plays the "synthesizer" role in an AI Council run — combining the top-scoring attempts into a final answer. Unset uses the Council's built-in default. Text; accepted values: string (≤120).


Cron failure alerts always fire

When on, cron-job failure alerts bypass Focus Mode and fire immediately regardless of your focus settings. Off by default, meaning they respect Focus Mode like other alerts. Turn it on if a permanently failed scheduled job is important enough that you never want Focus Mode to delay the notification. Toggle; accepted values: true | false.


Cron failure alerts

When on (the default), Omniscio shows an OS toast and creates a persistent inbox card whenever a scheduled cron job fails permanently. Turn it off to silence those notifications entirely, though you will then need to check job status manually. Toggle; accepted values: true | false.


Custom prompt rules

Your list of per-model or per-provider custom instruction rules. Each rule targets one or more engine families or specific model IDs; any session whose provider or model matches gets all the matching rules' text injected as extra instructions at the start of the conversation. Empty by default, meaning no extra instructions are added. Manage rules in Settings → Sessions → Custom Prompt Rules. List of objects; each object has id, targets (array), and text.


Custom CSS

Turns your custom stylesheet on or off. Custom CSS lets you write your own style rules for the whole app and see them apply straight away, for example to change colors or spacing. Nothing changes until you write a stylesheet. When this is off, the stylesheet is not applied, so you can switch it back on later. Omniscio also leaves it off automatically when it starts in Safe Mode, so a stylesheet that broke the screen is out of the way. The Save & Apply and Undo buttons in the same card work on the stylesheet itself. You find it in Settings > Appearance, in Advanced appearance on the Theme & Visual tab.

Default: on.


Allow custom provider sessions

The master switch for custom providers: the AI services you add yourself, meaning any service that uses the same connection format as OpenAI or Anthropic, plus the built-in OpenRouter entry. When it is on, they can be picked for new sessions. When it is off, none of them can be launched, though your saved providers are not removed. You find it in Settings > Accounts, in the Custom Providers card.

Default: on.


Daily digest sources

Which data sources are pulled into your Daily Digest briefing. Each source (calendar, Gmail, RSS, Slack, sessions, and others) can be toggled independently. Turn off sources that are too noisy or not relevant to your workflow to keep the digest focused. Object with boolean fields for each source.


Daily journal time

The wall-clock time (as a string, e.g. "20:00") of the daily journal nudge: the Journal's one gentle nudge a day while the Journal is on (set in Settings → Plugins → Journal), and AI Coaching's daily journal reminder while that feature is on. It is read in the time zone shown under the controls: changing the time saves the zone your device is in at that moment together with it, and if your device later moves to another zone the nudge stays on the saved one until you press Use my current time zone. Works together with dailyJournalWeeklyDay (for AI Coaching's weekly review). Text; accepted values: string.


Daily journal time zone

The time zone your daily journal check-in time and weekly review day are read in. Omniscio saves your computer's time zone the first time it needs one, and again whenever you change the nudge time or the weekly review day. If your computer later moves to another zone, for example when you travel, the schedule stays on the saved zone until you press Use my current time zone. The zone in use is shown as a line of text under the journal controls. Needs AI Coaching and the Daily Journal check-in to be turned on. You find it in Settings > AI Coaching.

Default: blank until first used, then your computer's time zone.


Daily journal weekly review day

Which day of the week your weekly review nudge fires. Uses JavaScript's day numbering: 0 = Sunday, 1 = Monday, …, 6 = Saturday. The day is read in the same time zone as the nudge time, and changing it saves your device's current zone the same way changing the time does. Requires the Daily Journal to be enabled. Number; accepted values: integer 0–6.


Default agent instructions (custom text)

The custom instructions text that is injected into every new session when the default agent instructions feature is enabled. Use this to set standing context — coding standards, your preferred response style, project conventions — that you want every session to know from the start. Up to 10,000 characters. Text; accepted values: string (≤10,000).


Default agent instructions

Master switch that injects your custom instructions text into every new session. When on, the content of defaultAgentInstructionsCustom is prepended to each session's context. On by default — write your instructions in Settings → Agent Instructions. Turn it off to stop the injection without deleting the text. Toggle; accepted values: true | false.


Default projects folder

The file-system path Omniscio suggests first when you create a new project and need to pick a working directory. Set this to your usual development workspace so you spend less time navigating the folder picker. Text; accepted values: string (≤1,000).


DeepSeek balance warning lead time

How much warning the DeepSeek balance forecast gives you before the projected run-out, in hours. Default 24. Raise it toward the 720 (30-day) maximum to hear about a slow drain much earlier; set it to 0 to turn the forecast steps off while the balance history keeps being recorded. The "DeepSeek is out of credit" notice is not a forecast — it still appears if the account actually reads zero. Only relevant when deepSeekBalanceMonitorEnabled is on. Number; accepted values: 0–720.

Deep-Target Highlight

Lets an agent, or a shareable link, take you straight to a page in Omniscio and then spotlight one specific control on it. For example, it can jump to a Settings section and point at the exact toggle to change. It only moves you to a page and points at something: it never starts a session and never changes a setting for you. This feature is part of the app and has no switch; an old saved value for this key is ignored.


Demo Look

Makes Omniscio look like a fresh install in one click: it saves your current look and presentation state, then puts a clean default appearance on screen, so you can show the app without your own theme, layout or panel arrangement visible. It persists, so the demo look survives an app restart until you turn it off — and turning it off restores exactly the look that was saved the moment you switched it on. Off by default. It is a toggle on the Settings home screen. Toggle; accepted values: true | false. The saved look is held in demoLookSnapshot, which the app manages and you never edit directly.


App icon attention badge

Shows a count badge on the Omniscio taskbar / dock icon (a Windows overlay icon, or the macOS / Linux dock badge). The count matches your Inbox count — every inbox item, not just sessions — and shows 0 while a surface is silenced. On by default. Turn it off for a quieter desktop if you prefer to check the inbox on your own schedule. Toggle; accepted values: true | false.


Dev Pipeline archive guard

Refuses to archive a session while any worktree it owns still holds unshipped work — not tagged ready-to-merge, not landed on master, and with no pull request carrying it — so a branch of finished-but-unlanded work cannot be closed away by accident. Every worktree the session owns is checked, not just the one recorded on the session: an agent that commits in a second folder it holds (an apply wave's apply/… branch, say) is held back too. Work counts as shipped however it actually got there: a tag, an in-app land record, an ordinary merge, a squash merge, or a pull request whose head carries those commits whatever that PR's branch is called. A refused archive returns the reason to the agent that asked for it, and never becomes an approval card in your sidebar. When you archive a session yourself (the sidebar X, a middle-click, the panel's archive button, the right-click menu, Manage Sessions, a Weekly Summary suggestion, the E key, the inbox's multi-select or right-click Dismiss, or by voice) and the guard refuses, the error toast shows its full reason, naming the branch and what clears it, instead of a generic "Couldn't archive" line. The session stays fully alive instead, and that toast offers Archive anyway: press it to archive the session regardless, with the guard left on for everything else. Only a person can do that; an agent or automation stays held. Voice and Weekly Summary archives show the reason without the button. Fail-open: if the check itself cannot run, the archive is allowed through. Off by default. Toggled from Settings → Features → Dev Pipeline options. Toggle; accepted values: true | false.

The guard also recognises work that was re-published as a fresh open or merged pull request on a newer copy of master, under a different branch name, even when that pull request's commits are not on your computer. It reads the pull request's changes from GitHub and counts the session's work as shipped when every file the session changed is in that pull request with the same changed lines. A pull request that left out part of the work, changed the session's own lines, or only shares blank lines or lone brackets with it does not count, so that session stays held. It looks at your 100 newest pull requests.


Custom pipeline phases

Turns on the Dev Pipeline's custom-phase editor, so you can add phases of your own (and their approval gates) at the point in the flow where they belong, alongside the built-in ones — each custom gate then gets its own auto-approve switch at its true position. Your phases are stored in ~/.claude/dev-pipeline/phases.json, which is the one file both Omniscio and the Claude skill read, so a phase you define is visible to the run it configures. Off by default. Toggle; accepted values: true | false.


Dev Pipeline

The main on/off switch for the Dev Pipeline, a guided workflow for software changes with five approval stops along the way. It is meant for people who build software with Omniscio. Turning it on installs the /dev-pipeline and /ready-to-merge skills into your Claude Code skills and switches on the automatic merge of finished work, the session reminder and the daily housekeeping jobs. Turning it off removes the skills and switches those off, though a merge helper you armed on its own stays on for the repos you picked. The same switch is on the Setup tab of the Dev Pipeline panel, where you also choose which repos it covers and how hands-on you want to be. You find it in Settings > Features.

Default: off.


Dev Pipeline repositories

Which repositories the Dev Pipeline — and through it the auto-lander — is on for, one entry per project. An empty list is not "off": with no entries every repository is on, including a repository you add later. An entry turns that ONE repository on or off — it is a per-repository override, never a whitelist, so listing a repository never narrows the Dev Pipeline to that repository alone, and a repository you leave off the list stays on. To stop the Dev Pipeline in one repository, turn it off on the Dev Pipeline panel's Setup tab — that toggle is the switch, and leaving a repository out of the list does not stop it. While the Dev Pipeline is on, this selection is also what the auto-lander watches; with the Dev Pipeline off, landing follows the auto-lander's own repository list instead. Nothing is written to your settings on your behalf, so there is nothing to undo. List of objects; accepted values: array of object { enabled: true | false; projectId: string (≤100) }.


Dev Pipeline sidebar item

Whether the Dev Pipeline panel's row appears in the sidebar. On by default. This is a visibility toggle and nothing more: hiding the row does not turn off the controls inside the panel and does not stop the background jobs behind it. Settings → Sidebar. Toggle; accepted values: true | false.


Diff viewer font size

Font size (in pixels) used for code content in the diff viewer and code-review panel. Range is 10–20. Increase it if the diff text is hard to read at a glance; decrease it to see more context on screen at once. Number; accepted values: integer 10–20.


Diff viewer tab width

How many spaces wide a tab character appears in the diff viewer. Range is 2–8; default behavior matches your code-review panel. Adjust to match the convention of the projects you review most often. Number; accepted values: integer 2–8.


Distribute sessions across accounts

When on, new sessions are spread across all your logged-in Claude accounts by usage load rather than always filling the first account first. This keeps costs and rate limits balanced across accounts. Off by default since 2026-08-29 — fresh installs run a single account, and multi-account users opt in here (the toggle appears with 2+ login accounts); installs from before that date keep the setting they had. When off, new sessions always go to your primary account. Toggle; accepted values: true | false.


Drip campaigns

Enables the Drip feature — a workflow that sends a sequence of scheduled messages (a "drip") to a recipient over time. Off by default. Use it to automate follow-ups, onboarding sequences, or spaced reminders. Toggle; accepted values: true | false.


Pause all drips

A global kill switch for Drip: when on, every active drip sequence is paused immediately and no scheduled messages go out. Off by default. Use this when you need to stop all drip activity at once — for example, during a product incident — without deleting individual drips. Toggle; accepted values: true | false.


Render HTML drip files live

When on, HTML files attached to a drip step are rendered as a live preview rather than shown as raw markup. On by default. Turn it off if you would rather see the raw markup. Toggle; accepted values: true | false.


Keep the documentation library current

Keeps the bundled documentation library in step with the published docs site. On by default: the app quietly checks for a newer published build and adopts it once it is verified whole, and the copy shipped inside the app stays in use whenever that cannot happen — offline, blocked, or a broken download. Turning it off stops the check entirely and reads the built-in copy alone. Toggle; accepted values: true | false.


ElevenLabs voice

The ElevenLabs voice ID used for text-to-speech playback when ElevenLabs is your selected TTS provider. Set the ID of any voice from your ElevenLabs account. Leave blank to use the provider's default. Text; accepted values: string (≤100).


Email inbound approved senders

The one list of outside senders you trust, kept per channel (SMS, Slack, Telegram and email). The Email Inbound card edits its email row: those addresses may file bug reports by email, take the fast path past the bug-report screen, and start paid automation work from an authenticated email. An empty email row keeps bug intake closed unless an Inbox Secret Token is set, and leaves email automations open; an empty SMS, Slack or Telegram row lets nobody start paid work. Map; accepted values: channel name to a list of sender ids or email addresses.


Agent email inbox

Enables the agent email inbox — an address (provided by Omniscio) that people can email to start or continue an agent session. Off by default. Once on, configure the inbox ID and approved senders. Toggle; accepted values: true | false.


Agent email inbox ID

The ID of the email inbox associated with your agent email address. This is set automatically when you configure the agent email feature and is used to route arriving email to the right session. Text; accepted values: string (≤200).


Email inbound pre-screen prompt

Instructions given to the pre-screener agent when an email arrives from a sender who is not on the approved list. The screener uses this prompt to decide whether to accept the message, ask a clarifying question, or reject it. Customize this to match your desired intake policy. Text; accepted values: string (≤16,000).


Archive summarized emails

When on, emails are automatically archived after the Email Summarizer processes them. Off by default — emails stay in your inbox after summarization. Turn it on if you prefer a clean inbox where only un-processed messages remain. Toggle; accepted values: true | false.


Email summarizer daily cost cap

The maximum amount (in US dollars) the Email Summarizer is allowed to spend on AI calls in a single day. Default is $1.00; range is $0.10–$20. Raise it if summarization is being cut off because you receive a high volume of email; lower it to keep costs in check. Number; accepted values: 0.1–20.


Email summarizer default prompt

The prompt template the Email Summarizer uses when generating a summary for each email. Leave blank to use the built-in default template. Edit it to change the summary style — for example, to always include action items or to use bullet points. Text; accepted values: string (≤20,000).


Email summarizer

Enables the Email Summarizer, which reads each incoming email and generates a short summary that appears alongside the message in your inbox. Off by default; requires an email inbox configured. Toggle; accepted values: true | false.


Summarize Gmail emails

When on, the Email Summarizer also processes emails from your connected Gmail account in addition to the agent email inbox. Off by default; requires both the Email Summarizer and Google authentication to be set up. Toggle; accepted values: true | false.


Email summarizer inbox ID

Identifies which inbox the Email Summarizer reads. Set automatically when you configure the feature. Text; accepted values: string.


Enable Cron Jobs

Shows or hides the Cron Jobs entry in the sidebar. Cron Jobs is where you schedule tasks to repeat on their own. Turn this off if you never use scheduled tasks and want a tidier sidebar. It only hides the entry: jobs you have already set up keep running while it is hidden. You find it in Settings > Features.

Default: on.


Enable Recipes

Shows or hides the Recipes entry in the sidebar, and the Recipes option in a project's right-click menu. Recipes are saved multi-step plans you can run again. Turn this off if you do not use them. Your saved recipe files are left untouched, so turning it back on brings everything back. You find it in Settings > Features.

Default: on.


Enable Skills integration

Shows or hides the Skills entry in the sidebar. Turning it off also stops Omniscio from checking whether your installed skills are taking up too much room, so you will not get that warning card in your inbox. Nothing is uninstalled, and the skills themselves are not changed. You find it in Settings > Features.

Default: on.


Engineering principles prompt

When on (the default), every regular session receives a short system-prompt note with senior-engineer discipline: keep solutions simple and scoped, fix the root cause instead of papering over symptoms, treat reported bugs as real, write genuine tests, and stay alert to performance and security issues. Skipped on Ask Omniscio helper sessions. Turn it off in Settings → Features if you find the added instructions interfere with specialized tasks. Toggle; accepted values: true | false.


Execution bias prompt

When on (the default), every regular session gets a short system-prompt nudge to act on an actionable request in the same turn rather than just describing a plan, to keep going to a genuine finish rather than stopping early, and to retry a weak or empty result a different way. It still asks when it needs you and never overrides an explicit stop or approval gate. Skipped on Ask Omniscio helper sessions. Turn it off in Settings → AI Coach. Toggle; accepted values: true | false.


Fathom API key

Lets Omniscio read your Fathom meeting transcripts so the "What should I do next?" card in Mission Control can use them, and only after you say yes to each use. To get a key, open Fathom and create one under Settings > API Access, then paste it here. It is stored encrypted on this computer, and once saved you can replace or remove it. When a key is saved, the Fathom connection for agent sessions can use it too, and the Fathom row of the MCP server list shows an "Add your Fathom key" link until a key is saved. You find it in two places that share the same one saved key: Settings > My API Keys (the Meeting notetakers group) and, in Settings > Mission Control, the Meeting notetakers card in the Omni area. A key saved or removed in one place shows the same in the other. Saving it on My API Keys does not by itself let What should I do next use it; the Mission Control card still offers a "Use my saved key" button for that. That feature is still in development, so the field may not be visible on every install yet.

Default: no key saved.


Trust Claude Code's own final answer over the agent's marker

A sub-option of "Show only the final message". Claude Code tells Omniscio where an agent's real answer begins. With this on, that wins over the spot the agent marked itself. This helps when an agent writes something for you, such as a draft email, a document or a list, and then puts its marker in the wrong place. That could hide the text and leave you a reply that says "see above" about something you cannot see. The trade-off is that on some replies you will also see the agent's working notes above its report. It is an experiment, so it is off until you try it. It is grayed out while "Show only the final message" is off. You find it in Settings > Features.

Default: off.


Always hide final message marker

When on, the final-message marker is hidden from the chat view in all circumstances, even in debug situations where it would normally be visible. Advanced toggle; most users will not need this. On by default. Toggle; accepted values: true | false.


Final message marker debug mode

When on, extra debug information about the final-message marker detection is surfaced in the chat. Useful for diagnosing why a session's final message is or is not being identified correctly. Off by default; intended for developers and advanced users. Toggle; accepted values: true | false.


Show only the final message

Teaches each new session to mark where its final answer begins. Omniscio then shows you just that final message and tucks the step-by-step work into a collapsible activity row, so long runs read as a short answer instead of a wall of steps. Nothing is deleted, and you can open the row any time. If a session does not use the marker, you see the usual full view. Because the instruction is given when a session starts, it applies to sessions you start after changing the switch. The options listed under it in Settings > Features are grayed out while this is off.

Default: on.


Show all blocks in final message

When on, all message blocks — including tool calls and thinking steps — are shown even when "Show only the final message" is active. Normally those intermediate steps are hidden to keep the view clean. Turn this off if you would rather keep the view clean and see the final message only. On by default. Toggle; accepted values: true | false.


Final message structural floor

A sub-option of "Show only the final message." When on, the final-message marker can hide tool calls and thinking, but it can never hide the actual answer or any live question the model wrote after its last tool step. This fixes non-Claude models such as DeepSeek, Codex, and GLM that sometimes place the marker in the wrong position, which would otherwise bury the answer you need to act on. Single-step turns without tool use are not affected. On by default. Toggle; accepted values: true | false.


Fireflies API key

Lets Omniscio read your Fireflies meeting transcripts so the "What should I do next?" card in Mission Control can use them, and only after you say yes to each use. To get a key, open Fireflies and create one under Integrations > Fireflies API, then paste it here. It is stored encrypted on this computer, and once saved you can replace or remove it. Agents you ask about your Fireflies meetings use the same saved key. You find it in two places that share the same one saved key: Settings > My API Keys (the Meeting notetakers group) and, in Settings > Mission Control, the Meeting notetakers card in the Omni area. A key saved or removed in one place shows the same in the other. Saving it on My API Keys does not by itself let What should I do next use it; the Mission Control card still offers a "Use my saved key" button for that. That feature is still in development, so the field may not be visible on every install yet.

Default: no key saved.


Enforce the fleet safety governor

The fleet safety governor watches how hard your computer is working and slows down background agent work when the machine is struggling, so the app stays responsive. It only paces agents, never something you started yourself. This switch decides whether it actually slows anything. With it on, background agent work is held back when needed. With it off, the governor keeps watching and measuring but never holds anything back, which lets you see what it would do without it doing anything. It has no effect if the governor itself has been turned off. A change takes effect straight away, with no restart. You find it in Settings > Diagnostics.

Default: on.


Run friction fixes under an Overseer

Hands each friction triage round to this project's Overseer instead of starting the triage session directly. The Overseer reads the ranked list of reported problems, decides which ones are worth fixing, and starts the triage helper and its fixers itself. Off by default. If no Overseer is running — or the one that is there is busy — the loop starts the triage session itself, so a round is never skipped. Toggle; accepted values: true | false.


Favourite models

The models you star in Settings → Sessions, under "Favourite models". When an agent starts a session it reads this list along with every model it may pick, so your favourites are the ones it reaches for first. Starring a model never restricts anything — a model you have not starred stays exactly as usable, the list is never reordered, and a favourite naming a model that has since been retired is simply ignored. Empty by default, so nothing changes for anyone who never stars one. Select the models you want favoured; the list holds model ids (≤200 entries).


Focus Mode

Enables the Focus Mode feature, which suppresses non-critical notifications and reduces visual interruptions while you need to concentrate. When off, the Focus Mode controls are hidden throughout the app. Toggle; accepted values: true | false.


Gemini virtual project

Adds a dedicated Gemini virtual project to the sidebar so you can run Google Gemini sessions alongside Claude sessions. Off by default; requires a Gemini API key configured in Settings → Accounts. Toggle; accepted values: true | false.


Gemini voice ID

The voice Google Gemini speaks with when Gemini TTS is selected as your Read Aloud provider. Choose a prebuilt voice from the picker in Settings → Voice Control, or type any other Gemini prebuilt voice name (they are capitalised, for example Kore or Puck). Defaults to Kore. Gemini TTS runs on your own Google API key — the same key the app's other Google features use — so there is no separate credential to enter here.


Git guardrails enforcement level

Controls how strictly Omniscio's git safety rules are enforced in agent sessions. required (the default) hard-blocks a guarded operation — the agent cannot proceed. advisory logs a warning and lets the operation through. Lower to advisory only if you need to allow an operation that the guardrails would otherwise block and you are sure it is safe. Select; accepted values: "advisory" | "required".


Git guardrails protected branches

The list of branch names that agent sessions are not allowed to commit to, push to, or edit files on directly. Defaults to ["master", "main"]. Add any other branches (for example a release or production branch) that should be treated as protected. List; accepted values: array of strings.


Require ready-to-merge tag before push

When on, the git guardrails also require a current SHA-bound ready-to-merge tag on a feature branch before an agent can push it. This enforces the full ready-to-merge workflow (lint, typecheck, tests, and build must pass) for every push. Off by default because it imposes the workflow; opt in per team preference. Toggle; accepted values: true | false.


Let an agent unblock itself briefly

When on (the default), an agent that a git guardrail has just blocked may open a short, bounded self-pause window to carry on — up to 10 minutes, and no more than 3 times an hour. It is meant for the case where you have already told the agent to go ahead and it would otherwise have to stop and ask you the same question again. Every use posts an inbox notice, so you always learn it happened and what it did. The window cannot lift everything: it never allows pushing to a remote, and it expires on its own rather than staying open. Turn this off to require your explicit approval every single time an agent hits a guardrail — with it off, only you can lift a block. Toggle; accepted values: true | false.


Glass dialogs

Gives pop-up windows a frosted glass look with a soft glow in your accent color. It follows your theme, light or dark. Turn it off for the classic flat look. This is a look-only choice and does not change how any dialog works. You find it in Settings > Appearance.

Default: on.


Frosted panel blur

Blurs the colors behind the two side columns (Hubs and sessions) and the top bar. It is off by default: in the glass themes the only thing behind those surfaces is the theme's still, already-soft color wash, so the blur changes nothing you can see while still costing graphics work. With it off those surfaces keep their tint, borders and shadow. Pop-up glass that floats over moving content (the mobile frosted header, the Vault side drawer, mind-map cards) keeps its blur either way. Turn it on if the columns look wrong without it. You find it in Settings > Appearance, beside Glass dialogs; pop-out windows follow it too.

Default: off.


Invite-only sign-in gate

When on, new users can only sign in to this Omniscio install if they have received an invitation. This is a server-enforced policy (the cloud enforces it for all users); the local setting mirrors the state of the server toggle. Managed by administrators only; changing it writes a policy to the server. Toggle; accepted values: true | false.


Global hotkey

The system-wide keyboard shortcut that brings Omniscio to the foreground from any app. Set it in Settings → System by pressing a key combination. Supports multiple bindings (stored as an array); the first binding is shown as the primary shortcut. Accepts an Electron accelerator string or array of strings (≤100 characters each).


Gmail default inbox query

The Gmail search query used when the Gmail panel first loads, determining which messages appear by default. Default is is:inbox. Change it to any valid Gmail search syntax — for example is:unread or label:important — to have a different starting view. Text; accepted values: string (≤500).


Export to Google Docs

Enables the Google Docs export feature, which lets sessions write their output directly to a Google Doc. Off by default; requires Google authentication. Configure the destination folder and optional template after enabling. Toggle; accepted values: true | false.


Google Docs destination folder

The Google Drive folder ID where exported Google Docs are saved. Paste the ID from the folder's URL. Leave blank to save to the root of your Drive. Text; accepted values: string.


Google Docs template

The ID of a Google Doc to use as a template when creating new exported documents. The new document is created as a copy of this template. Leave blank to create plain, untitled documents. Text; accepted values: string.


Google Workspace tools

Retired (2026-09-19). No Omniscio-built MCP server composes into a session any more, so the Google Workspace MCP toggle is gone. Google access is unaffected — it still runs through the gws/gog CLI backend (googleCliBackend) and the composed google-drive MCP server.


Grandfather existing mobile devices

When on (the default), mobile devices that were already paired before the device-approval feature shipped are automatically treated as approved and keep working without requiring a manual approval step. Flip it off to force those grandfathered devices through the full pending → approve flow — useful for testing or enforcing approval for all devices going forward. Explicit approvals are not affected. Toggle; accepted values: true | false.


Grok voice

The voice ID used for text-to-speech playback when Grok (xAI) is your selected TTS provider. Set the ID of a preset Grok voice. Leave blank to use the provider's default. Text; accepted values: string (≤100).


Run the local CLI control server

The control server is the local address (127.0.0.1:19519) that lets outside scripts and agent sessions drive Omniscio — spawning sessions, changing settings, reading your data. On by default. Turn it off to stop it binding at launch; the change takes effect the next time Omniscio starts. While it is off, command-line and agent tooling that talks to Omniscio cannot connect, including the other switches on this page which govern its routes. It is the user-facing twin of the AMC_DISABLE_CLI_SERVER=1 operator knob: either one stops the server.


Devin API key

The service-user key that Devin's REST API issues, used when you run a session on Devin's cloud instead of this computer. Paste the key into the Devin card under Settings > Accounts; it is stored encrypted and never shown back in full. Leave it blank to keep Devin unavailable as a session provider.


Grok Bot backup webhook key

The secret the Grok Bot backup webhook signs its calls with. It is used only when Omniscio cannot reach the Grok Bot app through its saved sign-in and falls back to the backup webhook URL. Paste the key into the Grok Bot card under Settings > Accounts; leave it blank if you do not run a backup webhook.