ui: Forty-Two chat shell v1

[bobbygodias]

18/08/2026 01:35 BRT — Frontend shell checkpoint

First branded frontend slice for Forty-Two AI.

This PR intentionally preserves the existing chat/session engine

• keeps ChatView, message rendering, keyboard behavior, drafts and input pipeline intact;
• replaces the predecessor-facing quick control with a Forty-Two-branded side panel;
• keeps requested backend separate from effective backend;
• makes the header status dot reflect the effective runtime instead of behaving like a generic online indicator;
• preserves CPU fallback and existing runtime store behavior.

02/09/2026 07:15 BRT — Launcher and real quick controls

Completed in this checkpoint

• installs the official Forty-Two Android launcher artwork in all legacy densities;
• adds Android 8+ adaptive launcher and round-icon resource definitions;
• preserves the 512 px Play Store derivative under assets/branding;
• shows the active Pal in the Forty-Two quick panel;
• opens the existing model/Pal picker from the quick panel;
• opens the existing generation-parameter and preset sheet from the quick panel;
• routes the legacy header menu and the quick panel through one shared settings sheet;
• records the active branch, source-bundle hash and verification limits in PROJECT_CANON.md.

Verified locally

• TypeScript compilation passed;
• repository lint passed with zero errors;
• localization and bundled-font checks passed;
• 13 focused header/menu routing tests passed;
• PNG dimensions, XML parsing and Android resource-variant inventory passed.

The local Gradle wrapper could not download Gradle 9.0.0 because that isolated environment could not reach services.gradle.org. The GitHub Android workflow is the authoritative native build gate.

Deliberate boundaries

• no Kotlin/JNI bridge changes in this checkpoint;
• no claim that Vulkan inference is active from hardware detection alone;
• runtime-orchestrator work remains isolated in draft PR runtime: introduce Forty-Two runtime orchestrator foundation #1 until the two lines are reconciled deliberately.

Next UI work

• splash/intro integration;
• Models + Lab surfaces;
• Joy UX/microcopy layer;
• device validation of requested versus effective backend/offload before runtime integration.

Draft until CI and device-level build gates pass.

[bobbygodias]

Android Test-04 checkpoint — 2026-09-02

• Commit: b76752199e1de0e7c25810d57e969852714aa836
• Workflow: Forty-Two AI Android CI #35
• Run: https://github.com/bobbygodias/Forty-Two-AI/actions/runs/33618821341
• Result: passed
• Built variant: prodDebug, ARM64, MEGA 3 native target set, Vulkan device-count probe
• Artifact: forty-two-ai-mega3-vulkan-test-04-device-count-probe
• APK handoff name: Forty-Two-AI-MEGA3-Vulkan-Test-04-UI-Icons-QuickControls-b767521.apk
• APK size: 233337596 bytes
• APK SHA-256: 0895bb8ba13cb3c34112bd431cd73cfa652f7a736ff7becd195640fe19d61480
• Packaging checks: ZIP integrity passed; launcher icon resources present in all Android densities plus adaptive/round XML; ARM64 librnllama.so packaged; APK Signing Block present.
• Signing state: debug-signed (prodDebug), for personal/device testing only.
• Real-device validation: still pending. A successful CI build does not prove Vulkan inference on Bobby's hardware.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b76752199e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Report the actual GPU backend instead of Vulkan

When a user selects GPU/hybrid on iOS or the stock Android build, these labels report Vulkan even though setRequestedBackend() merely selects the generic gpuBackendDevice; getDeviceOptions() identifies that device as Metal on iOS and OpenCL on Android (src/utils/deviceSelection.ts). Because the same labels are presented as both requested and effective backend, benchmarkers can record Vulkan execution that never occurred. Derive the displayed backend from the selected device/native telemetry, or retain the generic GPU label until Vulkan execution is verified.

Useful? React with 👍 / 👎.