|
Download ROADMAP.md from Abrarul8541/ModularAI-Terminal: direct link, hf CLI and curl.
- Browser
- Download file 2.98 kB
-
https://huggingface.co/Abrarul8541/ModularAI-Terminal/resolve/main/ROADMAP.md
- Command line
-
hf download hf://Abrarul8541/ModularAI-Terminal/ROADMAP.md
-
curl -L -o ROADMAP.md https://huggingface.co/Abrarul8541/ModularAI-Terminal/resolve/main/ROADMAP.md
2.98 kB
| # Roadmap (original 5 phases + missing pieces filled in) | |
| Legend: [ ] todo, [~] in progress, [x] done and tested on a real device | |
| ## Phase 0 - Foundations (was missing) | |
| - [~] Repo, license, README, architecture doc | |
| - [ ] Tech stack decision: Kotlin + Jetpack Compose UI, Room (SQLCipher), native llama.cpp via JNI (NDK, arm64-v8a only to hold size down) | |
| - [ ] CI: GitHub Actions / HF Jobs build, lint, unit tests, signed release APK, R8 shrink + resource shrink | |
| - [ ] Size budget gate: CI fails if base APK > 15 MB | |
| - [ ] Crash reporting (local only, no telemetry), logging | |
| ## Phase 1 - UI / UX | |
| - [ ] Universal Capture Bar (text, voice, share-intent, clipboard) | |
| - [ ] Dashboard + cards, glassmorphism and flat themes, dark/light/system | |
| - [ ] Responsive layouts (phone, tablet, foldable), accessibility (TalkBack, font scale) | |
| - [ ] GUI <-> raw CLI instant toggle | |
| - [ ] Onboarding + permission explainer screens (missing in original) | |
| ## Phase 2 - Core architecture and storage | |
| - [ ] Room database with SQLCipher | |
| - [ ] AES-256-GCM vault, key in Android Keystore, optional biometric unlock | |
| - [ ] Backup / export / import (encrypted), wipe-all | |
| - [ ] Settings + migration framework (missing in original) | |
| ## Phase 3 - Model station | |
| - [ ] Device profiler: SoC, ABI, RAM total/free, GPU (Vulkan/OpenCL), storage speed | |
| - [ ] Tier matrix: Ultra-Light 1-3B, Balanced 7-8B, Power 14B+ -> quant, ctx, est. tokens/s | |
| - [ ] Downloader: HuggingFace, Ollama registry, direct URL; resume, checksum (SHA-256), Wi-Fi-only, disk-space check | |
| - [ ] GGUF loader (llama.cpp), ONNX Runtime backend, cloud fallbacks (OpenAI, DeepSeek, Anthropic) with keys in vault | |
| - [ ] Hot-swappable storage dir (mount / unmount / wipe), user override for oversize models | |
| - [ ] On-demand localization: fetch 20-50 KB JSON locale, hot-swap, persist or purge | |
| ## Phase 4 - Power modes and plugin sandbox | |
| - [ ] 7-tap hidden dev menu | |
| - [ ] Foreground service with wake-lock management and battery-optimization guidance | |
| - [ ] PRoot rootfs installer (Ubuntu/Alpine/Debian), terminal emulator, apt/apk/pip | |
| - [ ] App functions exposed as CLI commands | |
| - [ ] Termux bridge (RUN_COMMAND intent), ADB-over-wifi, optional root/su | |
| - [ ] Plugin runtime: manifest, declared permissions, sandboxed execution, event hooks, signature check | |
| - [ ] Accessibility / overlay services, off by default, explicit consent (Play-policy aware) | |
| ## Phase 5 - Testing and updates | |
| - [ ] Unit + instrumented tests, benchmark suite for tokens/s | |
| - [ ] In-app updater: signed update manifest, SHA-256 + signature verification (replaces unsafe "hot reload" in release builds) | |
| - [ ] Debug-only local OTA / hot-reload server | |
| - [ ] Test-build distribution, changelog, versioning | |
| - [ ] Release checklist: APK attached only after install + smoke test on device | |
| ## Notes on gaps in the original roadmap | |
| Security model, CI, size budget, updater signing, migrations, accessibility, onboarding and legal/licensing of downloaded models were absent and are added above. | |