# Performance settings These settings control how much work Omniscio does behind the scenes and how it shares your computer with everything else. Most of them exist to answer one question: **when the app starts feeling slow, what is worth changing?** Two rules before you dive in: - **Nothing here changes what your sessions produce.** Every switch on this page affects only how Omniscio draws itself and how it schedules work — never the answers your agents give. - **Start with the recommended ones.** Anything labelled recommended is on by default because it measurably helps. The rest are opt-in experiments or Windows-specific tuning for particular machines. If the app feels slow right now, the three worth trying first are **Lite mode**, **Low power mode** and **Show only your messages and the agent's replies**. --- ## Lite mode (optimize for a less powerful computer) {#liteMode} One switch that makes Omniscio lighter on an older or slower computer. Turning it on enables Low Power Mode, holds new sessions back when memory is low, reclaims memory from idle sessions on Windows, staggers how sessions load, and stops keeping extra chats pre-built in the background. Turn it off to go straight back to the settings you had before. Anything to do with startup takes effect the next time you open Omniscio. Accepted values: on / off. Off by default. --- ## Keep my computer smooth under load (recommended) {#smoothLoadEnabled} When you have a lot of sessions running at once, Omniscio eases off its own background monitoring so your computer stays responsive. It does not change anything your sessions produce — only how hard Omniscio works behind the scenes. Turn it off for the older, always-on monitoring. Accepted values: on / off. On by default. --- ## Speed up agent commands (skip a safety check that cannot apply) {#guardPreScreenEnabled} Skips a slow security check for agent commands that could not possibly trigger it, which makes each command noticeably cheaper. Anything involving git or your master branch still gets the full check, completely unchanged. Omniscio installs and manages the hook for you and keeps a backup, so you can turn this off again at any time. On computers that do not have that heavy check installed, it does nothing at all. Accepted values: on / off. Off by default. --- ## Run agent hooks in a resident host {#hookHostEnabled} Every agent tool call runs a few small safety and housekeeping checks. With this on, Omniscio keeps one long-lived helper process running them, so each call starts one tiny native program instead of a full Node runtime — hundreds of fewer process starts a minute on a busy machine, which is what keeps the app smooth when several sessions are working at once. Turning it off restores the old one-process-per-hook behaviour at the next session start; nothing about what your sessions do changes either way. Windows only for now — other systems keep the per-call behaviour regardless. Accepted values: on / off. On by default. --- ## Run background jobs in one shared process {#backgroundRunnerEnabled} Omniscio runs scheduled background jobs on its own — health checks, clean-ups, watchers. With this on, they all run inside one long-lived helper process, so a job no longer starts a whole new program every time it runs — hundreds fewer program starts an hour. A job that gets stuck is stopped at its time limit without holding up the others, and everything the helper started ends when Omniscio closes. Turning it off runs each job in a program of its own again from its next run; which jobs run, and when, stays exactly the same. Accepted values: on / off. On by default. --- ## GPU acceleration {#disableGpu} Uses your graphics card to draw the app. Leave this on unless you are seeing graphics glitches or the app will not start. Turned off, drawing falls back to the processor and the app will feel noticeably slower. Accepted values: on / off. On by default. --- ## Low power mode {#lowPowerMode} Turns off the visual extras — background blur, heavy animation and decorative shadows — so the app draws more smoothly on a laptop or a computer with built-in graphics. It takes effect straight away; no restart needed. Accepted values: on / off. Off by default. --- ## Fast session switching {#fastSessionSwitchingEnabled} Makes clicking between sessions feel snappier by not re-drawing the panels you are not switching to. It changes nothing about what your sessions do, and reverts instantly if you turn it off. Accepted values: on / off. On by default. --- ## Cheaper live updates when many sessions run {#perSessionSubscriptionIsolation} Each live update a busy session sends wakes only its own panel instead of nudging every open session. Worth trying if you often run many sessions at once and the app feels busy. It re-wires instantly with no restart, and never changes what your sessions do. Accepted values: on / off. Off by default. --- ## Smooth streaming output (recommended) {#streamingThrottleEnabled} Paints fast-streaming text about 30 times a second instead of on every single token. Long answers look smoother, and the rest of the app stays responsive while several sessions stream at once. Turning it off paints every token the moment it arrives — more processor work, and nothing is more correct for it. Accepted values: on / off. On by default. --- ## How often background sessions refresh {#streamingThrottleHiddenMs} How often text from sessions you are **not** looking at gets repainted, in milliseconds. A higher number means less background work when many sessions stream at once. The session you are actually viewing always stays smooth, and a background session catches up as soon as you switch to it. Only applies while smooth streaming output is on. Accepted values: 100 – 5000 milliseconds. Default 500. --- ## Show only your messages and the agent's replies {#realConversationLayoutEnabled} Hides the step-by-step tool work and folds each turn into a collapsed "N actions" pill you can open whenever you want the detail. This is the single biggest speed-up for busy sessions. Turn it off for the older view that shows every intermediate step as it happens. Accepted values: on / off. On by default. --- ## Keep recently viewed sessions instant {#frozenPanelRetentionEnabled} Keeps the sessions you have looked at recently fully built in memory — the way a browser keeps background tabs ready — so switching back to one is instant instead of a rebuild. A built-in limit caps how many are held so memory stays bounded, and only sessions that are not actively streaming are kept. Trades some memory for speed. Accepted values: on / off. Off by default. --- ## Keep The Vault loaded for instant switching {#kmsKeepWarmEnabled} Once you have opened The Vault, Omniscio keeps the editor loaded in the background so coming back to it is an instant flip instead of rebuilding it every time. Only one editor is held, and only after you first open The Vault. Desktop only. Accepted values: on / off. On by default. --- ## Keep recently opened panels loaded {#panelKeepWarmEnabled} Keeps the panels you have opened — AI Coaching, Calendar, Drive, Sheets, Supermail and Team Chat — loaded in the background, so switching back into one is instant instead of rebuilding and re-fetching it. Each panel is only held after its first open. Desktop only. Accepted values: on / off. On by default. --- ## Keep Tasks loaded for instant switching {#tasksV2KeepWarmEnabled} Once you have opened the Tasks panel, Omniscio keeps it loaded in the background so coming back to it is an instant flip instead of rebuilding the list every time. Desktop only. Accepted values: on / off. On by default. --- ## Show checkboxes on tasks {#tasksV2ShowCheckboxes} Shows a checkbox at the start of every task. Turn it off for a cleaner list with no checkboxes. You can still finish a task from the keyboard (press **x** or **e**), by middle-clicking the row, or from the row's menu. Accepted values: on / off. On by default. --- ## Checkboxes select tasks instead of finishing them {#tasksV2CheckboxSelects} Makes task checkboxes work like an email inbox: ticking a box selects the task rather than completing it, so a stray click can never clear a task by accident. A bar appears as soon as anything is ticked, with **Done** and **Dropped** on it. Holding Ctrl or Shift always selects, whichever way this is set. Accepted values: on / off. Off by default. --- ## Finish tasks without the confirmation step {#tasksV2SkipFinishConfirm} Removes the "are you sure?" prompt when you finish or drop tasks, so a batch clears in one click instead of two. **Undo** still appears every time, and finished tasks move to Archived rather than being deleted, so a mistake is always recoverable. Deleting a task always asks first, whatever this is set to. Accepted values: on / off. Off by default. --- ## Where new tasks are added {#tasksV2QuickAddPosition} Choose where a task you add with the Quick Add row appears in the list. **Top** puts new tasks above the others; **Bottom** adds them to the end. Accepted values: Top / Bottom. Default: Top. --- ## Compact task rows {#tasksV2CompactDensity} Tightens the spacing between task rows so more of them fit on screen. Applies both to the All Hubs view and to each Hub's task list. Accepted values: on / off. Off by default. --- ## Hide completed tasks {#tasksV2HideCompleted} Hides tasks you have finished wherever they still appear, mainly the Done section of the Today view. Most finished tasks leave the list on their own when you complete them, so this only affects the few that stay behind. Accepted values: on / off. Off by default. --- ## Pre-build inbox sessions {#prerenderInboxSessionsEnabled} Builds your most recent inbox (needs-you) sessions in the background so switching into them is instant. A built-in limit keeps the app fast even when the inbox is long; older ones take about half a second to open. Turn it off to do no pre-building at all. Accepted values: on / off. On by default. --- ## Keep chat responsive under heavy load {#leanHiddenPanels} When many sessions are open, Omniscio stops the ones you cannot see from doing work every time another session sends a message. Scrolling stays smooth and the chat always lands in the right place, however many sessions are running. Turn it off to build every hidden session in full, as before. Accepted values: on / off. On by default. --- ## Pre-load new sessions {#preloadSessionsEnabled} Keeps one blank session ready in each Hub so that starting a new one with **Ctrl+T** is instant. Turn it off and new sessions start on demand — slightly slower, but lighter on your computer. Accepted values: on / off. On by default. --- ## Stagger session loading at startup {#panelMountStaggerEnabled} Spreads out the building of background session panels after the app starts or reloads, instead of building them all at once, which can freeze the window for a few seconds on launch. The session you are looking at always loads first and immediately; the rest follow within a few seconds. Accepted values: on / off. Off by default. --- ## Reclaim RAM from idle sessions (Windows only) {#wsTrimIdleMinutes} Periodically asks Windows to release the memory held by Claude sessions that have been sitting idle for at least this many minutes. The sessions are not closed and nothing is lost — the memory is simply handed back and reloads in about a tenth of a second the next time the session is used. Set to 0 to switch this off. Accepted values: 0 – 1440 minutes (0 = off). Default 0 (off). --- ## Release idle sessions to reclaim more memory {#idleSessionReleaseMinutes} Once a session has finished its turn and been waiting on you for this many minutes, Omniscio fully releases its background engine — reclaiming all of the memory it was holding, roughly 250–450 MB for each session, rather than just trimming it. Your conversation stays on screen and is reloaded in the background the moment you send your next message, which makes that first reply slightly slower. A session that is waiting on your answer to a question or a plan keeps its engine. Set to 0 to switch this off. Works on every platform. Accepted values: 0 – 1440 minutes (0 = off). Default 15. --- ## Reclaim memory automatically when it runs low {#memoryReclaimEnabled} When your computer's free memory drops and stays low, Omniscio tidies up the memory its own idle sessions are holding. It never interrupts a session that is mid-reply, and it only ever touches Omniscio's own sessions. No effect on macOS or Linux. Accepted values: on / off. On by default. --- ## When to reclaim memory {#memoryReclaimFreePercent} The trigger point for the tidy-up above: once free memory stays below this share of your total, Omniscio starts reclaiming. A higher number means it acts sooner and more often; a lower number means it only steps in under real pressure. Accepted values: 1 – 90 percent of total memory. Default 15. --- ## Cap runaway processes in a single session (Windows only) {#sessionMaxProcessesEnabled} Puts a firm ceiling on how many programs one session can run at once, so a single misbehaving agent — a runaway loop, an over-parallel command — cannot flood your computer and freeze it. The limit is generous and sits far above anything a normal session reaches, so it only ever stops something genuinely out of control. Nothing already running is ever killed; new programs simply wait until some finish. Accepted values: on / off. On by default. Windows only. --- ## Cap combined CPU use when many sessions run (Windows only) {#sessionCpuCapEnabled} Puts a ceiling on how much processor power all Omniscio Claude sessions together may use — sessions, subagents, test runs and builds included — once your computer is busy. A single session you are waiting on can still use the whole machine; the ceiling only applies when something else wants the processor. Omniscio also runs sessions at below-normal priority so they give way to your own apps. Accepted values: on / off. Off by default. Windows only. --- ## Maximum combined CPU use {#sessionCpuCapPercent} The ceiling used by the setting above, as a share of your computer's total processor power. A lower number is friendlier to your other apps but slower for heavy batch work. Accepted values: 25 – 90 percent. Default 50. --- ## Never go above the CPU ceiling, even on an idle computer {#sessionCpuCapStrict} By default the ceiling is soft: it only applies when other apps want the processor, so a single session you are waiting on can still use the whole machine. Turn this on to make it a hard limit that applies all the time — a stronger guarantee, but a lone session will be slower. Accepted values: on / off. Off by default. Windows only. --- ## Give Omniscio's own window priority (Windows only) {#boostAmcProcessPriority} Runs Omniscio's own window above normal priority while sessions run below it, so the app — and other apps you are using, like your browser — stay responsive when many sessions are saturating the computer. Accepted values: on / off. Off by default. Windows only. --- ## Keep Omniscio's database in memory (Windows only) {#keepAmcResidentInMemory} Holds Omniscio's database in fast memory so Windows cannot push it out to disk and then have to read it back from a busy drive — which is what causes the occasional whole-app freeze during heavy multi-session work. It sizes itself to your computer, keeps within a safe limit so it cannot crowd out your other apps, and switches itself off on computers with little memory. Only Omniscio's own database is held, never your sessions. Accepted values: on / off. On by default. Windows only. --- ## Serve the database from private memory (Windows only) {#keepDbInPrivateCache} Serves Omniscio's database from a block of memory Windows cannot take back, instead of reading it from the file on disk. Under heavy disk load Windows can evict a file-based database, and the app freezes while it re-reads it; private memory cannot be taken back that way, so those freezes stop. Fully reversible — turn it off and restart to go back. Takes effect the next time you start Omniscio. Accepted values: on / off. Off by default. Windows only. --- ## Run the database on a background thread {#offThreadDbWorkerEnabled} Runs Omniscio's database reads on a background thread instead of the one that draws the window, so a slow disk cannot freeze the interface mid-query. This was long labelled experimental because an earlier version made multi-session use slower — database reads queued behind each other. The August 2026 rework of this worker cleared the failure that warning was about, so it no longer describes this build. It stays off by default. Switching it on or off takes effect immediately, with no restart. Accepted values: on / off. Off by default. --- ## Force the background database worker on (development builds only) {#dbWorkerAbForceOn} A measurement switch that only appears in development builds. It forces the background database worker on right now, bypassing its safety checks, so the two can be compared without opening a terminal. It is not a normal setting and is never on in the version you use. Accepted values: on / off. Off by default. Only present in development builds. --- ## Let sessions give way to your own apps (recommended) {#sessionTreeBelowNormalPriority} Runs every Claude session — and everything it starts, including searches, git, builds, tests and subagents — at below-normal priority, so it steps aside for the Omniscio window and your other apps when the computer is busy. A session you are waiting on by itself still uses the full machine; it only yields when something else needs the processor. This is the gentler, always-safe alternative to a hard CPU cap. Accepted values: on / off. On by default. Windows and macOS. --- ## Let sessions give way on the disk too (recommended) {#sessionTreeLowIoPriority} The disk side of the setting above. Sessions and everything they start run at low disk priority and below-normal memory priority, so when Omniscio needs to read from disk — or pull its own memory back in — it goes first. This is what keeps typing and clicking responsive while many sessions are hammering the drive. Sessions still run at full speed whenever Omniscio is idle. Windows only. Accepted values: on / off. On by default. Windows only. --- ## Pause all testing (temporary) {#pauseAllTesting} Stops any new test, typecheck, lint or build run from starting anywhere on this computer, so background testing stops piling up. Runs already under way finish normally — nothing is killed and no paid cloud run is thrown away. This is the switch to reach for when the machine feels bogged down by agent testing, because pausing your sessions does not stop it: a test run keeps going after the session that started it has stopped. It turns itself back on after about an hour, and restarting Omniscio turns it back on too, so testing can never be left off by accident. Accepted values: on / off. Off by default. --- ## Batch your agents' find commands (recommended) {#agentFindBatchingEnabled} Rewrites the file-searching commands your agents run so they work in efficient batches instead of starting one program per file. A single runaway search can otherwise start tens of thousands of programs and freeze your computer. It is also faster. Only turn it off if a specific command genuinely needs the old behavior. Applies to your next new session. Accepted values: on / off. On by default. --- ## Speed up merge batches {#mergePriorityBoostEnabled} While "merge all ready branches" runs, that job is given above-normal priority and your other sessions are briefly dropped to below-normal, so the batch finishes sooner on a busy computer. Your other sessions keep running — they just give way. It reverts on its own when the merge ends, after a 30-minute safety timeout, or as soon as you turn it off. Accepted values: on / off. Off by default. --- ## Spacing between sessions when many restart at once {#sessionRestartSpacingSeconds} When several sessions come back together — after an app restart, after recovering from a rate limit, or all at once from Manage sessions — this is how far apart Omniscio **spaces their starts**, so a big batch does not hit your computer all at the same time. It also keeps only a handful of them working at once, so a large restart cannot run every session concurrently and exhaust the machine. While the app is still opening it may start them sooner, but only once your computer has room. Sessions you start or restart yourself, one at a time, are never delayed. Set to 0 to fall back to the built-in spacing. Accepted values: 0 – 30 seconds. Default 1.5. --- ## Seconds between background starts during severe lag {#fleetSafetyGovernorSevereSpacingSeconds} How cautiously background agent work proceeds while Omniscio detects severe interface lag. A higher number gives stronger protection for your computer; a lower number lets agents move faster. Normal operation, and anything you start yourself, is never delayed. Leave this alone unless you have a reason to change it — by default Omniscio works out the right value for your machine. Accepted values: seconds. Left unset by default, so Omniscio derives the right value for this computer. --- ## Auto-land ready branches during a merge batch {#autoLanderEnabled} While "merge all ready branches" runs, a background worker also lands branches the moment they are tagged and merge cleanly, so the main merge only has to stop for genuine conflicts. It never touches a branch somebody is resolving by hand, it only ever moves things forward so nothing can be lost, and it stays entirely on your computer — it never pushes anywhere. Only active where the merge tools live. Accepted values: on / off. Off by default. --- ## Bring my local master up to date automatically {#masterAutoSyncEnabled} Lets a background check bring your local master branch up to date with the remote by itself — about five minutes after the app starts, then every six hours. Leave it off and you are still **told** when you have fallen behind, with a one-click catch-up; nothing moves your branch unless you ask. Only applies when you are running Omniscio from source, and it never runs while the auto-lander is on. Accepted values: on / off. Off by default. --- ## Run automatic worktree cleanup {#worktreeCleanupEnabled} Turns on scheduled, load-aware cleanup that reclaims disk space by retiring finished git worktrees. It also brings up a dashboard with the history of runs, a plain explanation of why each worktree was kept or removed, one-click recovery if one was removed by mistake, and a "run cleanup now" button. Accepted values: on / off. Off by default. Windows only. --- ## Announce when a worktree is cleaned up {#showWorktreeCleanupToasts} When you archive a session that already has a pull request, its worktree is cleaned up automatically. This shows a brief notification when that happens. Turn it off for fully silent cleanup. Accepted values: on / off. On by default. --- ## Push sessions onto efficiency cores when the CPU is busy (Windows, hybrid CPUs) {#sessionAdaptiveEcoreAffinity} On computers with a mix of fast and efficient processor cores (Intel 12th-generation and newer, including Core Ultra), this keeps Claude sessions on the efficient cores while overall processor use is above about 85%, and lets them back onto every core once it drops below about 60%. Your sessions keep working while the fast cores stay free for Omniscio and your browser. It does nothing on computers that do not have that mix of cores, so it is safe to leave on. Accepted values: on / off. Off by default. Windows, hybrid CPUs only. --- ## Ease memory pressure by paging busy sessions to disk (Windows only) {#sessionAdaptiveMemoryPaging} When your computer is low on memory, Omniscio tells busy sessions — and the tests they run — to keep less in fast memory and put the rest onto the disk. Sessions keep running; they simply slow down until memory frees up. It starts above roughly 85% memory use and lifts again below roughly 70%. It never affects Omniscio itself, and never cancels, blocks or kills anything. Accepted values: on / off. Off by default. Windows only. --- ## Hold new sessions back when memory is low {#sessionMemoryGuardEnabled} When your computer is low on memory, this pauses new sessions that arrive in a burst — automatically created bug and feature sessions, recipe runs, mass restarts — until memory frees up, so a flood of sessions cannot push the machine into a freeze. A single session you open by hand is never delayed, and anything held starts within a few seconds. Sessions resuming after an app restart are always paced this way regardless of this setting. Works on every platform. Accepted values: on / off. Off by default. --- ## The memory level that triggers holding sessions back {#sessionMemoryGuardPercent} The point at which the setting above starts holding burst sessions back, as a share of your computer's total memory in use. Lower is more cautious and holds sooner; higher means it only steps in when memory is nearly full. Accepted values: 50 – 95 percent. Default 85. --- ## Wait between freshly created sessions {#cliSpawnSpacingSeconds} How long to wait between starting brand-new sessions that an agent or automation creates in a burst — a script opening many sessions in a row, for example. This stops rapid automated starts from piling up all at once. Sessions you open yourself are never affected. Set to 0 to start them immediately. If your computer is very busy, a waiting session can be held back briefly to protect it, but never more than 30 seconds past its turn, so at the default setting the queue still starts one every 30 seconds. Accepted values: 0 – 300 seconds. Default 30. --- ## Stop sessions loading tools they do not use {#sessionStrictMcpConfig} Every session normally starts all the tool servers listed in your global Claude configuration, including ones it never touches — the Playwright browser tool alone can take about 95 MB each. With many sessions open, those idle servers add up to gigabytes of wasted memory and can slow the whole computer down. With this on, each session loads only the tools Omniscio actually set up for it, so the waste cannot build up, and sessions start a little faster. Your Omniscio tools — memory, The Vault, Google and any custom servers — keep working exactly as before. Applies to Claude sessions; Codex, Gemini and Cursor are unaffected. Accepted values: on / off. On by default. --- ## Keep the computer awake while sessions are running {#keepAwakeWhileSessionsRunning} Stops your computer going to sleep while one or more sessions are running, so long jobs keep working when you step away. The screen can still sleep to save power — only system sleep is blocked — and normal sleep returns the moment every session has finished. Works on Windows and macOS; on Linux it depends on your desktop environment. Accepted values: on / off. On by default. --- ## Run the active session in its own process (experimental) {#isolatedSessionViewEnabled} Draws the session you are viewing in its own background process inside the main window, so it stays responsive while the rest of the app is busy. It takes effect the next time you click a session. Accepted values: on / off. Off by default.