EdgeDox – Offline document AI on Android using Qwen3.5-0.8B

Name: EdgeDox – Offline document AI on Android using Qwen3.5-0.8B
Availability: InStock
Author: cyberfly-labs

by cyberfly-labs·Mar 7, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemDark Horse

Offline RAG on mid-range Android, but Ollama + local LLM apps already exist.

Strengths

•Local RAG implementation on ARM CPUs with quantized Qwen3.5 avoids cloud dependency
•Handles PDFs, TXT, and Markdown with document context retrieval
•Early beta with real optimizations (MNN inference, memory profiling for devices)

Weaknesses

•Context window severely limited by mobile RAM; CPU-only latency likely painful
•Nascent adoption (50+ downloads); unclear if model quality sufficient for real workflows
•Local RAG quality depends entirely on embedding model—no clarity on choice or fine-tuning

Post Description

Hi HN,

I’ve been experimenting with running small language models directly on mobile devices and built a small Android app called EdgeDox.

The idea was to make document AI usable without sending files to a cloud service. Many existing tools require uploading PDFs or documents to a server, which can be a privacy concern.

EdgeDox runs a lightweight language model (Qwen3.5-0.8B) locally on the device so documents stay on the phone.

Current features:

• Ask questions about PDFs • Document summarization • Extract key points from long documents • Works completely offline • No accounts or server processing

The model runs locally using mobile inference (MNN). I'm experimenting with quantized models and other optimizations to keep memory usage and latency reasonable on mid-range Android devices.

Some challenges so far:

• balancing context size with mobile memory limits • improving response latency on CPU-only devices • reducing model load time

The project is still in early beta, and I’m mainly looking for feedback from people experimenting with on-device AI or mobile inference.

Play Store: https://play.google.com/store/apps/details?id=io.cyberfly.ed...