We have Omarchy running natively on my AYN Odin 2 Portal, and all the work is public! Nothing held back, PRs welcome, or take it and make it your own. 🎮🐧 Booting from internal UFS storage. Android userspace removed. No VM, no Android dual boot, and no microSD card involved in the installation. The project is omarchy-mobile-odin2 (github.com/TechLuddite/omarc…). Before the details: huge credit to @dhh (xitter.cf/dhh) and the Omarchy contributors, @simonschubert (xitter.cf/simonschubert) for omarchy-mobile/moarchy and the mobile keyboard, and the ROCKNIX and Pocknix communities for the handheld Linux work this builds on. Their work made this possible. Credits are right near the top of the README because they belong there. The goal is a usable Omarchy handheld: a native desktop, touch controls, configurable gamepad shortcuts, a game library, and eventually the wider Odin hardware working properly. This has been a hands-on, AI-assisted bring-up. I’m working with Codex, applying the evidence-and-verification approach from my SeeFix project: inspect the actual device, save the logs, check the hashes, use the camera when software access disappears, and distinguish a device that enumerates from hardware that actually works. We started with USB, the hardware buttons, and recovery/bootloader access. No SD card to take the usual removable-media route. Before replacing anything, we used Qualcomm EDL to capture the original base system, boot firmware and platform/calibration data. This is a base backup, rather than a clone of personal userdata. Seven image extents totaling 11.57 GiB, compressed to about 2.7 GiB; SHA256 checks, both GPT copies across all six UFS LUNs, 115 partition entries, and independent rereads of both ABL partitions and persist all passed. Restoration itself is still untested. Then came a Portal-specific Linux RAM bootstrap, USB networking and SSH, followed by the native Arch Linux ARM/Omarchy filesystem. We replaced Android userdata with a 512 MiB FAT32 boot volume and an ext4 root, preserving the preceding firmware/calibration partition entries. The former Android super partition became a read-only recovery filesystem with a second verified base-backup copy and native recovery images. The off-device backup remains the primary copy. The installed stack uses pinned Omarchy 4.0.3, Simon’s mobile adaptations, the ROCKNIX Portal kernel/device tree/modules/firmware, and ROCKNIX ABL 1.1.8. Host-side build tooling uses Docker/QEMU; the handheld itself boots its own Linux kernel and runs the ARM userspace directly. Getting a desktop on screen was only the beginning. The bootloader menu offered USB, SD and Auto, rather than the “Internal” option described elsewhere. Auto worked, but we also had to explicitly select AYN Odin 2 Portal in the loader’s device-model menu. Otherwise: “device model not set.” The screen initially came up upside down. Correcting display rotation still left touchscreen gestures rotated: a sideways swipe opened a panel that should respond to a vertical swipe. Matching the touchscreen transform to the display fixed that. The on-screen keyboard opens and dismisses, and I joined Wi-Fi directly from the handheld. A Qt 6.12 Color-name collision also made the pinned mobile shell’s bar disappear. A compatibility adjustment restored its colors, clock, network and battery indicators. Graphics are native too: EGL reports Freedreno FD740 and Vulkan reports Turnip Adreno 740. The panel is running at 120 Hz, rotated to landscape. Actual hardware rendering on the Portal. One of the biggest usability lessons was the gamepad mapper. A “press the keyboard shortcut you want” workflow is awkward when the device has no physical keyboard. So we adapted @pinku1 (xitter.cf/pinku1)’s Controller Control into a touch-focused Omarchy Gamepad Mapping fork: record a button or combo, pick an action from the live Omarchy shortcut catalog, then save. With more than 200 actions, that picker quickly needed search too. The mapper passes 52 tests on the ARM handheld, including 44 upstream tests. Home is assigned to Terminal. Rear paddles and the broader physical-controller checklist still need verification. The fork is public and linked from the main repo: omarchy-gamepad-mapping (github.com/TechLuddite/omarc…) @btsouth (xitter.cf/btsouth)’s Omakade 1.16.1 is also installed from the signed ARM64 repository and opens in the native Wayland desktop. Its optional guide overlay needs compatibility work with our pinned shell, so that part is disabled. Launching games is still a separate acceptance step. We also hit a serious boot-recovery hurdle after changes to the redundant boot chain. A microphone experiment’s reboot exposed it, but the loader rejected the unchanged, previously working boot image too. That mattered: the failure couldn’t simply be attributed to the new microphone routes. Stock slot-A AVB metadata still described the Android system we had removed. Replacing it with reviewed native metadata for the unlocked loader restored image acceptance. Then we hit a bug in our own recovery-image builder: FAT formatting on a host file defaulted to 512-byte sectors, while the Portal’s UFS exposes 4096-byte logical sectors. Linux refused to mount /boot. We recovered through the saved RAM bootstrap, read the installed journal, rebuilt the FAT image with 4096-byte sectors, and verified the full device readback after repair. Normal internal boot now reaches the desktop again. /boot mounts, Hyprland and the mobile shell run, USB SSH works, and systemd reports no failed services. Upstream qbootctl marked the verified slot A boot successful; both GPT copies across all six LUNs passed the follow-up checks, with only the expected success flags and checksums changed. We’ve also configured persistent, key-only user and root SSH over Wi-Fi, with local hostname discovery. Both logins passed while USB networking was disconnected in software, and root file-transfer round trips passed. Development and maintenance no longer depend on keeping the USB cable attached. The microphone is the main unresolved hardware problem. @peteonrails (xitter.cf/peteonrails)’ native ARM64 Voxtype build is installed, its release signature was verified, and a known speech file transcribes correctly in both the CLI and our touch-operated Microphone Check app. Live internal-microphone tests still produce no usable words. The strongest lead is the WCD codec’s digital-microphone routing over SoundWire. We have experimental device-tree/UCM work to investigate, but it is rolled back from the current working configuration. Back → dictation waits until actual voice capture passes. Noise processing can come afterward; first we need a working input signal. Cold power-cycle acceptance, alternate-slot recovery, audible speaker/headphone tests, Bluetooth pairing, USB peripherals, paddles, haptics, fan/LED behavior and real suspend/wake still need work. This is a documented bring-up checkpoint. The installation tooling is specific to the captured Portal layout. More shout-outs to the people and projects behind the README: @viperbjk (xitter.cf/viperbjk) for EDL tooling; @armbian (xitter.cf/armbian), Armada and the Renegade Project for related native Odin installation work; Caleb Connolly, linux-msm, @linuxfoundation (xitter.cf/linuxfoundation), @linaroorg (xitter.cf/linaroorg) and @collabora (xitter.cf/collabora) for Qualcomm boot/recovery work and guidance; AOSP for AVB tooling; @lcMenci (xitter.cf/lcMenci), @docker (xitter.cf/docker) and QEMU for the ARM build foundations. @Ayn_Loki (xitter.cf/Ayn_Loki) and the AYN UCM contributors, Teguh Sobirin, Philippe Simons and Luke Johnson for board, input and audio work. ROCKNIX and Shuuri Labs/Pocknix deserve another mention here too. And the foundations underneath all of it: Arch Linux ARM, Linux/V4L2, Mesa/Freedreno/Turnip, ALSA, @PipeWireP (xitter.cf/PipeWireP) and WirePlumber, @hyprwm (xitter.cf/hyprwm), @outfoxxedd (xitter.cf/outfoxxedd) and Quickshell, @qtproject (xitter.cf/qtproject), @GTKtoolkit (xitter.cf/GTKtoolkit)/PyGObject, @FFmpeg (xitter.cf/FFmpeg), BusyBox, OpenSSH, NetworkManager, Avahi, dosfstools and mtools. The main repo keeps the source revisions, original notices, installation/recovery tools, hardware checklist, failures and remaining gaps together. The mapper has its own fork and credits. Device images, credentials and raw captures stay private. I wanted Omarchy on the handheld I already own. Now it boots natively, the desktop is usable by touch, and we have a recovery record to build from. Next: work through the hardware, one verified result at a time. #Omarchy #Linux #AYNOdin2 #OpenSource