Lite mode (optimize Omniscio for a less powerful computer)

What it is

Settings → Performance → Lite mode. A single switch that makes Omniscio lighter on low-spec ("less powerful") machines by flipping a curated bundle of existing, individually-tested Performance settings at once. It adds no new performance engine — it is a one-click shortcut over knobs that were already there but scattered across ~20 separate toggles a non-technical user would never find and set by hand. Off by default. Fully reversible: turning it off restores the exact settings you had before.

Engineering invariants + the tests that lock them: lite-mode-preset-contract.md.

Where to find it

Settings → Performance → Lite mode. It also offers itself to you: on a machine it judges low-spec, an offer card explains what it would change before you accept. Turning it back off restores exactly the settings you had before, so it is never a one-way door.

How it behaves

What it turns on

When Lite mode is ON it writes these settings (plain-language effect):

Setting (key)	Lite value	Effect
Low Power Mode (lowPowerMode)	on	Drops GPU-expensive visuals (background blur, heavy animations, decorative shadows).
Keep recent sessions instant (frozenPanelRetentionEnabled)	off	Stops holding extra recently-viewed chats fully built in memory (saves RAM).
Pre-render inbox sessions (prerenderInboxSessionsEnabled)	off	Stops pre-building inbox chats in the background (saves CPU/RAM; they load on click).
Stagger session loading at startup (panelMountStaggerEnabled)	on	Spreads panel building over a few seconds after launch instead of freezing on startup.
Hold new sessions when memory is low (sessionMemoryGuardEnabled)	on	Pauses a burst of new sessions when RAM is tight so the machine can't be driven into a freeze.
Reclaim RAM from idle sessions (wsTrimIdleMinutes)	5	(Windows only) Evicts idle Claude CLI processes' RAM after 5 min; pages reload on next use.

These are the same settings documented individually in Settings → Performance — Lite mode just sets them together. Startup-related changes (stagger) take effect on the next launch.

The one-click offer on the weak-hardware cards

Omniscio detects weak hardware at startup (low RAM, integrated GPU, few CPU cores — detectWeakHardware). A computer below the recommended specs gets the below-spec warning card; one whose weakness is only its graphics gets the "Your computer may benefit from Lite mode" card. Both carry the same one-click "Turn on Lite mode" button: it turns on the whole bundle, confirms with a "Lite mode is on" message, and archives the card — a failed save leaves the card in place so you can try again. Each card is shown at most once per install (the graphics card via liteModeOfferSeen) and is an offer, never an auto-apply — weak hardware never silently changes your settings.

The offer is gated on liteModeOfferSeen, so installs that already dismissed the older single-toggle low-power prompt still get this improved offer exactly once.

"Weak hardware" means only physical properties of the machine — RAM, cores, GPU model. Omniscio's own rendering switch is never part of it: hardware acceleration being off (a Safe Mode boot, AMC_DISABLE_GPU=1, the crash auto-disable, or the user's disableGpu setting) says nothing about how powerful the computer is. It used to count, which told a 24-core / 137 GB workstation it "looks less powerful" the one time it booted Safe Mode — and burned the once-per-install offer on that false alarm. Acceleration being off also blanks the GPU probe (gpuRenderer reads unknown), so detectWeakHardware skips the integrated-GPU check entirely in that state rather than guessing from a value it cannot see.

Reversible by design

• Turning Lite on snapshots the current value of each bundled setting into liteModeSnapshot (only on the off→on edge), then writes the lite values.
• Turning it off restores each setting to the snapshotted value and clears the snapshot. You get back exactly what you had — Lite mode is not a one-way door.
• It is not a computed master switch: it writes the real underlying settings, so each one keeps working through its own normal machinery.

What it deliberately leaves out

The bundle is conflict-free and high-confidence. It excludes:

• the hard CPU cap (sessionCpuCapEnabled) — it conflicts with "Speed up batch merges" and can throttle a single session you're waiting on; the always-on "let sessions yield to your foreground apps" is the gentler substitute;
• adaptive memory paging / e-core affinity — niche, Windows-hybrid-CPU only;
• message virtualization — superseded by the default-on real-conversation layout, which already cuts long-chat render cost, so forcing it would do nothing.

For agents

Where it lives

• Bundle + apply logic: src/shared/lite-mode.ts (LITE_MODE_BUNDLE, computeLiteModePatch).
• Settings keys: liteMode, liteModeOfferSeen, liteModeSnapshot in src/shared/types/settings/system-settings.ts.
• The toggle + the weak-hardware offer: src/renderer/src/features/settings/sections/performance/PerformanceSettings.tsx / src/main/services/system-health/hardware-detector.ts.
• The one-click card button: APPLY_LITE_MODE_ACTION in src/shared/alert-actions/performance.ts, run by the apply-lite-mode handler in src/renderer/src/features/alerts/use-alert-actions.ts.

Related

The individually-settable Performance options Lite mode bundles are described under the app's Performance settings, and the visual-only trimming it sits beside is Low Power Mode.