Memory footprint#

The motivation for Omagma was a single Gmail tab observed using roughly 2 GB of RAM on its author's desktop. That is an individual observation, not a universal Gmail requirement or a controlled comparison: the bar shows a bounded recent-mail list and snippets, and opens Gmail when needed. The terminal client can read full mail separately.

At a glance#

These are dated Linux x86_64 observations, not measurements of every later source change. The historical results and executable identities below remain attached to their original versions.

Bar totals exclude Chrome and unrelated shell widgets. Terminal totals exclude the emulator, external editor and separately running bar. Large bodies and a full cache have separate workloads below. Allocation limits describe reserved or capped storage; they are not a prediction of resident RAM. Installing Omagma does not require running these measurement procedures.

Published terminal workload, v0.2.6#

The published Linux x86_64 static-musl Safe executable, exact Zig 0.17.0, passed native CI with 100 warm-ups, 1,000 measured cycles and 60 quiet seconds. Its SHA-256 is 1d693cf3472857275396b6850a629fbb1c0bd3192b96be59d43f0bede2442a56; actual downloaded bytes separately passed functional checks and were installed.

All refused zero allocations and emitted no TUI output during their quiet minute. CLI, ordinary TUI and Markdown quiet CPU was zero; the large HTML workload observed 0.016666% of one core. Rounded decimal totals are about 9 MB CLI, 12 MB ordinary TUI, 13 MB Markdown composition and 31 MB for the large-body workload. These are separate lifetimes, not an isolated feature or compiler comparison. Release evidence records both Linux architectures and native Mac RSS/physical-footprint results. Older measurements and the dated bar estimate remain unchanged.

Published terminal workload, v0.2.5#

The published Linux x86_64 static-musl Safe executable, exact Zig 0.17.0, passed native CI workloads with 100 warm-ups, 1,000 measured cycles and 60 quiet seconds. Its SHA-256 is dbf57aca3c570f918135e72909921b17d474d2c5b24ab33db83923a696f58cc0; actual downloaded bytes separately passed functional checks and were installed.

All refused zero allocations. CLI quiet CPU was zero; both TUI workloads observed 0.016666% of one core and emitted zero quiet bytes. The cache observer uses a finite wait and leaves the idle screen unchanged. Allocation limits remain separate from these resident totals. Rounded decimal totals are approximately 8.5 MB CLI, 11 MB ordinary TUI and 29 MB for the large-body fallback. Release evidence records both Linux architectures and native Mac RSS/physical-footprint results separately. These process lifetimes do not establish an isolated compiler or feature cost. The dated live bar estimate above and earlier measurements below remain unchanged.

Published terminal workload, v0.2.4#

The actual published Linux x86_64 static-musl executable, using exact Zig 0.17.0 in Safe mode, completed 100 warm-ups, 1,000 measured cycles and 60 quiet seconds. Its SHA-256 is 5483e981701fb06b34f6a69d6f47b6c4fd236862ba87df4c031129553578dc03.

All three observed zero quiet CPU and no refused allocations. Both TUI workloads emitted zero quiet bytes. The application-owned heap stays capped at 64 MiB, with a separate 16 MiB backend/HTTP reservation; those bounds are separate from resident memory. The complete large-mail tail remains reachable.

Native Mac measurements use RSS and physical footprint rather than Linux PSS. Release qualification records their separate artifacts, platforms, workloads and downloaded/Homebrew checks. These synthetic process lifetimes do not replace the dated live three-account bar estimate or isolate a compiler effect.

Terminal modes, 0.2.2#

Cache-first runtime source c5d1473b996fcbec43e7405b103fb76fe104253b was measured on 2026-10-05 with exact Zig 0.17.0, native Linux x86_64 static-musl Safe. The original three-account workloads, 100 warm-ups, 1,000 measured cycles and 60-second quiet intervals retain their acceptance thresholds.

Rounded decimal footprints are about 9 MB for the CLI and 12 MB for the TUI, with workload peaks around 13–16 MB. Capped-heap peaks were 4,657,857 and 5,410,258 bytes, with zero allocation refusals. The TUI emitted no bytes during its quiet minute. These are whole terminal processes; the terminal emulator, external editor and separately running bar are excluded. Cache-first qualification records exact executable identities.

A separate maximum-cache stress seeded 2,000 metadata records averaging 1,528 bytes, added 100 and deleted ten. It retained the exact newest 2,000 within 3,344,073 disk bytes, with preserved body hashes and no orphan files. Its peak capped heap was 43,959,024 bytes and whole-process OS peak RSS 43,176 KiB, approximately 44 MB. This deliberately larger workload is separate from settled small-message memory; the 64 MiB heap ceiling and 16 MiB fixed-storage reservation are allocation accounting, not resident-memory totals.

The new terminal features settle roughly 2–4 MB above the previous local workload. Both versions use the same exact compiler, and their separate process lifetimes do not establish an isolated compiler effect. Historical tables stay attached to their original versions. The optional cache timer is a short native one-shot; it has no resident cache process between runs.

Terminal modes, 0.2.1#

The incoming-header compatibility patch at source 858ce32d21625a136fbed102d265e97bb95b0a48 was measured separately on 2026-10-05, using native Linux x86_64 static-musl Safe and exact Zig 0.17.0. The original three-account fixture workload, 100 warm-up cycles, 1,000 measured cycles and 60-second quiet intervals were unchanged.

Rounded decimal footprints are approximately 7 MB for the CLI and 8–9 MB for the TUI, with workload peaks around 11 MB and 12–13 MB. Terminal heap peaks were 6,692,585 and 7,619,937 bytes respectively, with no rejected allocations. The TUI emitted no output during its quiet minute. These are whole Omagma terminal processes; the bar, terminal emulator and external editor remain separate processes. The 64 MiB terminal heap ceiling and 16 MiB fixed backend/HTTP reservation are separate accounting bounds, already represented by their touched pages in process RSS/PSS. Patch qualification records exact executable identities.

The observed footprints remain close to the previous release below. These are separate process lifetimes, not an isolated compiler comparison; both terminal releases use the same exact Zig 0.17.0 compiler. Existing acceptance thresholds were unchanged.

Native v0.2.1 release CI run 37258436358 measured the same Safe workloads independently on Linux x86_64 and arm64 runners. Each used 100 warm-ups, 1,000 measured cycles and a 60-second quiet interval. The runtime source remains 858ce32d21625a136fbed102d265e97bb95b0a48; publication/tag revision b8d25227681090a9be0d8364330e8115bb672ffb records the release without changing it.

All four gates passed with zero private-memory growth and no rejected allocations; both TUIs emitted no quiet-window bytes. These are whole-process measurements on native CI hosts, with the same separate 64 MiB heap and 16 MiB fixed-storage accounting as the local results. Runner and process-lifetime differences affect high-water RSS, so the tables do not establish an architecture cost or an absolute RSS limit. The downloaded x86_64 release is byte-identical to the locally and natively qualified Safe executable and passed its own 45 CLI, 19 PTY and seven wire checks. No additional soak was needed for identical bytes. Terminal measurements do not include a desktop deployment or establish arm64 desktop behavior.

Preserved terminal modes, 0.2.0#

On 2026-10-05, native Linux x86_64 static-musl Safe builds from source 7568a2467b5a93142259bc50057a177c3acf7150 passed separate 1,000-cycle CLI and TUI measurements. Each used three fictional accounts with 96 messages each, after 100 warm-up cycles, followed by a 60-second quiet interval.

These numbers include the complete Omagma terminal process: touched storage, heap, stacks, runtime and executable pages. The terminal emulator, separately running bar, Chrome and an active external editor are separate processes. The CLI traversed eight pages and read a full message on every cycle; the TUI changed accounts, paged, selected messages and reloaded. Peak measurements and settled memory describe this workload, not a guarantee for every message or editor.

The terminal allocator peaked at 7,235,433 bytes for CLI and 7,617,129 bytes for TUI, with no rejected allocations. Its 64 MiB heap ceiling is separate from the 16 MiB fixed backend/HTTP reservation. The whole-process RSS/PSS measurements include their touched pages; summing allocation ceilings does not predict resident memory. See terminal qualification for reproduction and limits. Historical bar measurements below retain their original version and accounting.

Final mouse runtime, 0.2.3#

On 2026-10-05, source a94d29d07d2370d140bb13bef21840ead90b6c65 passed the combined local Linux x86_64 static-musl Safe gates with mouse enabled. Each ordinary CLI/TUI workload used 100 warmups, 1,000 cycles and 60 quiet seconds.

Both had zero quiet CPU and refused allocations; the TUI emitted no quiet bytes. The exact 2 MiB HTML fallback had 37,396 KiB process high-water, 56,049,942-byte heap peak, no warm growth, no refused allocations and zero quiet CPU/output. These measurements exclude the emulator, bar, Chrome and external editor. Final mouse evidence identifies the exact executables and workload. The pre-mouse checkpoint below remains unchanged.

Native release CI#

Release CI 37359163295 passed native x86_64 and arm64 terminal gates for the published 0.2.3 artifacts. Ordinary CLI/TUI settled at 9,060/11,624 KiB RSS on x86_64 and 6,336/8,828 KiB on arm64; all four had zero warm growth and quiet CPU, with no rejected allocations. Whole-process high-water marks were 13,124/14,792 KiB and 10,032/11,664 KiB respectively. The 2 MiB HTML fallback gates passed 1,000/60 on both platforms. Differences between runner/process lifetimes do not isolate architecture or compiler costs. Native ARM backend evidence does not qualify ARM desktop integration. Actual downloaded x86_64 bytes passed their own synthetic checks and were installed for that qualification; final evidence records release/deployment identity and preserved failures.

HTML-reader checkpoint, 0.2.3#

The pre-mouse source e4f852512646ce7cb4931677d391d3a49b82faeb passed separate Linux x86_64 static-musl Safe gates on 2026-10-05: 100 warmups, 1,000 measured cycles and 60 quiet seconds. These receipts precede the mouse implementation; its qualification remains separate.

All gates had zero rejected allocations and zero quiet CPU ticks; both TUIs emitted no quiet bytes. The boundary HTML exceeds the rich-layout line cap, so this case establishes complete plain fallback and reachable tail. It does not claim formatted rendering of all 2 MiB. The separate 2,000-message cache stress had 43,464 KiB OS peak RSS and 44,214,520-byte heap peak.

Rounded whole-process totals are about 9 MB CLI, 12 MB ordinary TUI and 29 MB settled for the boundary body, with respective high-water marks about 13/15/39 MB. These include touched storage, runtime and stacks, excluding the terminal emulator, bar, Chrome and external editor. The 64 MiB heap cap and 16 MiB fixed reservation are separate accounting; they must not be added to RSS/PSS. See the HTML checkpoint evidence for exact executable identities and preserved failures. Existing bar measurements below remain at their original versions.

Three-account bar measurement, v0.2.2#

On 2026-10-05, the installed downloaded v0.2.2 / Zig 0.17.0 static-musl backend with three connected accounts used 15,380 KiB PSS (15,392 KiB RSS), stable over five samples. The unchanged Quickshell UI's full offscreen lifecycle measurement attributed 14,433 KiB PSS after 1,000 cycles and 200 warm-ups. Together this estimates 29,813 KiB: 29.1 MiB / 30.5 MB, rounded to 31 MB for Omagma's backend and attributed UI share.

This is a combined estimate from separate live-backend and synthetic UI lifetimes, excluding Chrome and unrelated shell widgets. The UI receipt keeps its predecessor executable identity in v0.2.2 qualification; the component bytes are unchanged. The then-installed backend was the published x86_64 artifact, SHA-256 597009ad9dc3616c8f5473cc70f411ce280b7e640c86c50def0f6caaf02b712d. It was verified after rescan without replacing the desktop shell or account/profile settings.

The five-minute terminal-cache timer returns after each update and has no resident cache process between runs. An open TUI or a running cache one-shot is a separate process from the bar; terminal measurements above do not include the emulator/editor. Workload, process lifetime and Qt baseline differences explain why these observations do not isolate a compiler effect. Earlier version measurements below remain unchanged.

Three-account bar measurement, v0.2.1#

On 2026-10-05, the installed downloaded v0.2.1 / Zig 0.17.0 static-musl backend with three connected accounts and five-minute background refresh used 15,472 KiB PSS (15,480 KiB RSS), stable across five samples. A separate full offscreen UI qualification with 90 rows attributed 12,476 KiB PSS after 1,000 open/close cycles and 200 warm-ups. Together:

The README at that release rounded this estimate to 29 MB. It includes the complete backend and Omagma's attributed UI share, excluding Chrome and unrelated shell widgets. Combining a live backend sample with a separate synthetic UI baseline is an estimate, not a direct isolated desktop-process total. The actual installed executable SHA-256 is 4e936b367cdd6dd9ed5eac71ead732cb3b5f301cc047753ad8d5144b58b6e511. A separate downloaded-artifact live read-only test fetched 30 messages per account and passed a 60-second quiet interval with zero CPU ticks, added jobs or events; its backend PSS was 15,232 KiB. The plugin reload preserved the shell PID, private account/profile settings and closed popup.

The changed totals across versions include process-lifetime and UI-baseline variation. They do not isolate a compiler effect, and allocation ceilings still do not predict resident memory. Historical measurements remain attached to their actual versions.

Preserved three-account bar measurement, v0.1.1#

On 2026-10-05, the installed v0.1.1 / Zig 0.17.0 static-musl backend with three connected accounts and five-minute background refresh used 14,444 KiB PSS (14,452 KiB RSS), stable across five samples. A separate full offscreen UI qualification with 90 rows attributed 15,514 KiB PSS after 1,000 open/close cycles. Together:

The README at that release rounded this total to 31 MB. The then-installed executable was the actual downloaded x86_64 release, SHA-256 36d6c76847b44dbc4e4aac8255af75901ee6b3d17e6c20e1259a7a00c6d37e75. An independent read-only live test fetched 30 messages per account using existing grants and passed a 60-second manual-probe quiet interval. It measured 15,712 KiB backend PSS, illustrating variation between process lifetimes.

Preserved historical measurement#

On 2026-10-04, the installed static-musl v0.1.0 / Zig 0.16.0 backend with three connected accounts used 12,184 KiB PSS (12,192 KiB RSS). A separate synthetic UI lifecycle measurement with the production logo and 90 retained rows attributed 16,068 KiB PSS to Omagma over the Quickshell baseline. Together these give an approximate Omagma footprint of 28,252 KiB: 27.6 MiB, or 28.9 MB, roughly 30 MB.

That estimate includes the backend and Omagma's share of the UI. It is not the added cost of a second or third account. Omagma shares Quickshell with the rest of the bar, so UI attribution uses a baseline rather than assigning the whole desktop shell to one plugin. Chrome, other bar plugins and temporary keyring helper processes are excluded. The popup was closed after warm lifecycle use; opening it creates a temporary view.

RSS counts shared pages in full; PSS apportions them among processes. Combining RSS from the whole shell with the backend would describe the shell and its other widgets too. The application-owned 16 MiB storage reservation is a separate allocation bound, not the measured total process memory.

The historical release remains at 851ee30c5953a28fe0535fea03737ce85ec745fd. Private receipts are retained locally; only anonymized counts and memory totals are published. See historical evidence and reproduction and acceptance gates.

Zig 0.17 qualification#

The compiler port has separate dated evidence. Its native x86_64 Safe synthetic backend peak RSS was 11,624 KiB with 90 rows; the actual downloaded release's soak reached 11,636 KiB, the same peak as the historical 0.16 static release soak. Matched old/new UI diagnostics showed no compiler-linked increase. An earlier UI attribution gate failed and is preserved in the evidence: baseline initialization varies, so individual measurements are not guarantees.

The roughly 29 MB v0.1.0, 31 MB v0.1.1, 29 MB v0.2.1 and 31 MB v0.2.2 estimates come from different process lifetimes and UI baseline runs. They do not isolate the compiler's effect. Historical measurements remain attached to their actual version.