ofln™

Offline LLM chat for Android and iOS by Evolvyn AI (Darwin Chen). Download a GGUF model once, then run inference on-device via llama.cpp through llama.rn. No account and no cloud LLM for replies.

In the app: Settings → About for tips · Settings → Storage for backup.

Doc	Purpose
CONTRIBUTING.md	Dev build setup, code style, and PR guidelines
docs/PRIVACY.md	What stays local, Hugging Face, tokens, backups
SECURITY.md	Vulnerability reporting; Android signing files stay local
docs/TECHNICAL.md	Architecture, build/release, CI, Android 16KB, inference notes
DESIGN.md	UI system
docs/DEVICE_SMOKE.md	Maintainer device smoke checklist
TRADEMARK.md	Brand / fork rebranding policy

Screenshots

Curated Model Shelf	Chat & Ambient Hue	Personas Library

Scheduled Tasks	Performance Metrics	Perspective Debates

Features

• Browse / download GGUF models from Hugging Face (optional token for gated repos)
• Curated Start here shelf (Q4_0 phone-friendly picks) with SHA-256 integrity verification plus full HF browse / local import
• Download progress with percent and speed/ETA when measurable; resumable pause/resume
• Import local GGUF files; per-model settings (temperature, context, GPU layers)
• Streaming chat with optional reasoning/<think> parsing; calm per-turn tok/s
• Personas, temporary chats, pin/rename history
• Tasks (Settings): scheduled LLM prompts or Source Monitor (fetch URL + analyze) — results stay local
• Perspective debates: multiple on-device speakers take turns on a topic
• Image / PDF attachments (on-device OCR / text extraction; multimodal when supported)
• Voice input (speech-to-text) and spoken readout (text-to-speech)
• Light / dark theme
• Performance (Stages) metrics from on-device usage logs — private, not a leaderboard
• Backup & restore: chats JSON or full profile ZIP (models re-download via catalog; secrets never exported)
• Safe Mode recovery if a previous launch crashed during model load

Requirements

Requirement	Notes
Node.js ≥ 18
JDK 17+ (prefer 21)	Android builds (Gradle 8.12). Prefer Android Studio’s bundled JBR and set JAVA_HOME
Android Studio / SDK	Android
Xcode + CocoaPods	iOS (macOS)
New Architecture	Required by current llama.rn (enabled in project configs)

Setup

git clone https://github.com/Darwin27264/ofln.git
cd ofln
cp android/gradle.properties.example android/gradle.properties
npm install

iOS (macOS):

cd ios
bundle install
bundle exec pod install
cd ..

Run:

npm start          # Metro
npm run android    # Android device/emulator
npm run ios        # iOS simulator/device

No cloud LLM API keys are required. Hugging Face tokens (optional) are stored in Keychain and used only for huggingface.co downloads. See docs/PRIVACY.md.

Copy .env.example only if you add local tooling variables — the app runtime does not read a .env for inference.

Privacy (short)

Leaves device?	What
Only when you download/browse models	Hugging Face
Stays local	Chats, personas, Tasks results, Stages metrics
Never in backups	HF tokens, GGUF files

Inference does not use an ofln cloud API. There is no account system.

Backup & restore

UI: Settings → Storage

Action	Output
Export chats	ofln-chats-*.json
Export full backup	ofln-backup-*.zip
Import	Merge or replace

Full backups include chats, personas, perspective presets, tasks / recent runs, settings, and model catalog — not GGUF files or HF tokens. Models re-download on restore.

Development scripts

Script	Purpose
npm start	Metro bundler
npm run android / npm run ios	Build & run
npm test	Jest unit test suite
npm run lint:ci	ESLint (launch-critical paths)
npm run typecheck	Launch-critical path typecheck
npm run typecheck:full	Full-project tsc --noEmit
npm run android:build-release	Build release APK (assembleRelease)
npm run android:extract-release	Copy release APK into releaseAPK/ofln_{date}.apk

Direct Windows build & extract commands:

# Build release APK:
cd android; .\gradlew.bat assembleRelease; cd ..

# Extract to releaseAPK/:
powershell -ExecutionPolicy Bypass -File .\scripts\extract-release-apk.ps1

More build/release detail: docs/TECHNICAL.md.

Testing

npm run typecheck
npm run lint:ci
npm test

Unit tests live under __tests__/. Prefer pure helpers that do not require native modules. Manual smoke: docs/DEVICE_SMOKE.md.

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for local development setup, code style pointers, and PR guidelines:

• Prefer focused PRs; match existing TypeScript / React Native patterns.
• Visual or interaction changes: follow DESIGN.md and keep it current.
• New user-durable data: update backup export/import, schema/tests, and the Backup section above.
• Read docs/PRIVACY.md before claiming offline / zero-network behavior.

Support

ofln is free and open source, built for fun — if it saves you a download or a smile, a coffee is appreciated but never expected.

• GitHub Sponsors
• Ko-fi

License

GNU General Public License v3.0 — see LICENSE.

Copyright © 2026 Darwin Chen / Evolvyn AI.

Third-party libraries (including llama.cpp / llama.rn) retain their upstream licenses; GPLv3 applies to this project’s combined distribution terms.

Trademark

ofln™ and related branding are trademarks of Evolvyn AI. The GPLv3 license does not grant trademark rights. Forks that redistribute modified binaries must rebrand. See TRADEMARK.md.

Security

Report vulnerabilities privately — see SECURITY.md.

Android release signing stays in gitignored android/gradle.properties (template: android/gradle.properties.example). Do not commit upload keystores or passwords.