ofln v1.2 — Morning Brief, Translate from Any App, and On-Device Benchmarks

ofln is a free, open-source (GPLv3), offline AI assistant for Android that runs LLMs entirely on your phone with llama.cpp via llama.rn. No account, no cloud API, no telemetry: your prompts never leave the device.

v1.2 makes ofln useful every day, not just on airplanes: a scheduled Morning Brief, private Ask ofln actions from any app's text selection menu, a RAM-aware model picker, and a built-in tokens/sec benchmark.

---

✨ What's New in v1.2

☼ Morning Brief (scheduled, on-device summaries)

One tap on the Tasks screen sets up a daily brief: at 7:30 AM ofln fetches a page you choose (Hacker News by default), summarizes it locally, and can read it aloud. The fetch is the only network request; the summarizing happens on your phone.

✍️ Ask ofln from any app

Select text in any app and tap Ask ofln in Android's selection menu. Pick an action in the overlay:

• Summarize, Key points, Simplify, Rephrase
• Fix grammar, Tone (formal / casual), Draft a reply
• Translate (English ↔ Spanish)

Your result can replace the selected text in place, and it's all done offline.

🧠 RAM-aware model guidance

ofln checks your device's RAM and marks the best fit on the curated model shelf with ★ Recommended for your phone. Models that won't fit are hidden behind a "Show anyway" toggle, so there's no more guessing which quantization will crash.

⚡ Hardware acceleration benchmark

Measure real tokens/sec from the Performance screen on your phone's active backend: OpenCL GPU, Hexagon NPU (Snapdragon), or CPU. Copy the result as a Markdown card to share or compare.

📜 Context fullness & tap-to-summarize

When a long chat nears the model's context limit, a banner offers Summarize earlier turns to condense history on demand. A sliding window keeps your system prompt from ever being dropped.

🛡️ Privacy-safe diagnostic logs

Safe Mode exports now redact prompts and chat history before sharing a troubleshooting log. The separate Full Log button still copies the raw log for local debugging, so check it before you post it anywhere.

---

📥 Install (Android)

1. Download ofln-v1.2.apk from the Assets below.
2. Open it on your Android device. If prompted, allow "Install unknown apps" for your browser or file manager.
3. Launch ofln, download the recommended starter model, and chat fully offline.

Auto-updates: track this repo with Obtainium to get new releases straight from GitHub.

iOS: build from source (see the README).

---

📱 Requirements

• Android: 7.0+ (API 24+). Android 10+ recommended for modern OpenCL GPU drivers.
• APK: universal build bundling arm64-v8a, armeabi-v7a, x86, and x86_64 (16 KB page-size ready for Android 15+). A smaller arm64-only APK is planned.
• RAM: 4 GB minimum (1B–1.5B Q4 models); 6–8 GB+ recommended (3B–4B models).
• Storage: about 2–5 GB free for model weights.

---

🔒 Privacy

• Chats, personas, tasks, and usage stats stay on your device.
• The network is used only to download models (Hugging Face) and to fetch URLs you schedule in tasks such as Morning Brief.
• No accounts, no third-party trackers, no telemetry servers.

---

📦 Verification

Asset	Size	SHA-256
ofln-v1.2.apk	~212.9 MB	d80a966dde3d7865d37e2abce9e83fe50465c67ca40bf3257fe2b1a45cdda116

Verify with: sha256sum ofln-v1.2.apk

---

🤝 Contribute & Share

ofln is built in the open, and every bit of feedback shapes the roadmap.

• ⭐ Star the repo if you want offline AI on phones to keep getting better.
• 📊 Share your benchmark card: post your phone model and tok/s in Discussions or Issues to help build a real-world device table.
• 🐛 Report bugs or request features with the issue templates.
• 🔧 PRs welcome; see CONTRIBUTING.md.

Full Changelog: v1.1...v1.2

_{Keywords: offline AI app Android, local LLM on phone, private on-device ChatGPT alternative, llama.cpp Android, GGUF mobile, open-source AI assistant, F-Droid-friendly.}

ofln v1.1 — Initial Release

ofln v1.1 — Initial Release

Excited to introduce the first official public release of ofln™!

ofln is a private, completely offline, on-device AI assistant for mobile. Download a GGUF model once and run full conversational LLM inference directly on your phone using llama.cpp via llama.rn — no account, no cloud APIs, and zero data leaving your device.

---

✨ Key Features

• 🚀 100% On-Device Inference: Run quantized GGUF models directly on phone hardware with native C++ acceleration (OpenCL / Hexagon / CPU fallback).
• 📦 Curated Model Shelf: One-tap downloads for phone-friendly starter models (Q4_0 quants) with automatic SHA-256 integrity verification.
• 🌐 Hugging Face Browser & Local Import: Search and download any public or gated GGUF model directly from Hugging Face with resumable downloads, or import your own local .gguf files.
• 💬 Streaming Chat & Thinking Parser: Real-time token streaming with <think> tag parsing for reasoning models, plus quiet tokens/second telemetry.
• 📄 Multimodal OCR & Document Extraction: Built-in on-device OCR (ML Kit) for images and PDFs with intelligent RAM protection (temporarily frees model memory during OCR if RAM is constrained).
• 🎙️ Voice STT & Spoken Readout: Offline speech-to-text input and natural text-to-speech voice playback.
• 👥 Custom Personas & Ephemeral Chats: Create and swap custom agent personalities or start temporary privacy-first chats.
• 🤖 Scheduled Tasks & Perspective Debates: Automate background prompt runs, local web page analyses (Source Monitor), and multi-agent AI debates.
• 🛡️ Safe Boot Recovery: Automatic crash recovery and Safe Mode protection if memory pressure interrupts model loading.
• 🎨 Modern Ambient Interface: Fluid UI with dark/light themes, ambient lighting, and haptic feedback.

---

📥 Installation (Android)

1. Download ofln-release.apk below.
2. Open the downloaded file on your Android device.
3. If prompted, allow your browser or file manager permission to "Install unknown apps".
4. Launch ofln, browse the curated shelf, download your preferred starter model (e.g. Qwen 2.5 1.5B or Llama 3.2 1B/3B), and start chatting completely offline!

---

📱 Device Requirements

• Android: Android 10.0+ (API level 29+)
• Architecture: arm64-v8a (16 KB page-size ready for Android 15+)
• Memory (RAM):
  • Minimum: 4 GB RAM (supports 1B – 1.5B Q4 models)
  • Recommended: 6 GB – 8 GB+ RAM (supports 3B+ models)
• Storage: ~2 GB to 5 GB free storage for model weights.

---

🔒 Privacy Guarantee

• Your chats, custom personas, tasks, and usage metrics never leave your device.
• Internet connectivity is only used when explicitly downloading models from Hugging Face.
• No third-party trackers, no accounts, and no telemetry telemetry servers.

---

📦 Verification & Assets

Asset	Size	SHA-256 Checksum
ofln-release.apk	~212.8 MB	67fd16e3102594dc4076691e535571de2ca499a14641b4225b66a347bc8ded5f

---

📱 App Screenshots

Curated Model Shelf	Streaming Chat & Ambient Hue	Personas Library
Scheduled Tasks	Performance & Telemetry	Perspective Debates
:---:	:---:	:---:

Full Changelog: https://github.com/Darwin27264/ofln/commits/v1.1