KnoLo Core is a local-first knowledge base engine built for small language models (LLMs). It packages your documents into a compact .knolo file and enables fully deterministic querying — no embeddings, no vector databases, no cloud services required. Designed for on-device and edge LLM deployments.

On-device shell command generator for macOS Tahoe. Uses Apple's 3B model with dynamic few-shot retrieval from 21k tldr examples.

Agentic Android Open Source Project (AAOSP) — Android fork with native LLM system service, MCP-aware apps, and an agent-driven launcher. On-device Qwen 2.5 via llama.cpp. Apps declare tools in their manifest. The OS runs the model.

Android 16 fork. AI as a platform primitive. Twelve capabilities, one shared runtime, every app. OEM-pluggable. Apache 2.0.

High-performance Android SDK for on-device LLM inference (GGUF). Privacy-focused, offline-first, and powered by llama.cpp with a clean Kotlin Coroutines API.

Apple FoundationModels API on iOS 18+. Same call site, native passthrough on iOS 26 (Apple Intelligence), CoreML / MLX backends on older OSes. Drop-in source compatible.

📱 手机端 AI 操作系统全景知识库 — 334+ 篇深度页面，覆盖端侧大模型、AI Agent、芯片适配、推理优化 | 自动更新

iOS app that runs a local LLM on-device to transcribe meetings and generate structured notes — action items, decisions, and summaries. No cloud, no API keys, no data leaves the phone.

Ash — offline survival assistant for iOS. Gemma 4 E2B/E4B fully on-device (text · image · voice) with RAG-grounded answers over 56 emergency-response packs. Built for the Kaggle Gemma 4 Good Hackathon.

JibarOS organization profile.

온디바이스 LLM + RAG 기반 로컬 음악 추천 Android 앱 | Android local music recommendation app powered by on-device LLM and RAG

Execution infrastructure for local-first AI. Reason locally, execute globally.

On-device text-to-SQL distilled from GPT-4o-mini into Qwen2.5 (0.5B → 3B locally on M1 via mlx-lm LoRA, 7B+ on cloud A100). 847 MB at 62.5% on Spider dev; 3B variant hits 72.6%, 7B variant hits 75.0%.

Production KMP framework for Google LiteRT-LM. Run Gemma on-device with OEM-aware RAM fixes, resilient Ktor chunked downloads, and schema-driven function calling. Plain Android support. AGPL-3.0 / Commercial dual-licensed.

Run LLMs on Snapdragon NPU — including the 'unsupported' 8 Gen 1 (Hexagon v69). Verified at 31 tok/s on OnePlus 10 Pro.

Privacy-first macOS BCI prototype: Muse EEG → BrainFlow → Core ML (ANE) → local MLX LLM → SwiftUI carousel. Fully on-device.

Hands-free voice assistant for Android, fully on-device. Gemma 4 multimodal LLM via LiteRT-LM with optional ElevenLabs cloud TTS, smart-turn-classified barge-in, and a vision channel.

Documentation for MobileTransformers - a lightweight, modular framework based on ONNX Runtime for running and adapting large language models (LLMs) directly on mobile and edge devices. It supports on-device fine-tuning (PEFT), efficient inference, quantization, weight merging, and direct inference from merged models.

Compile Gemma 4 E2B for Snapdragon v69 NPU (8 Gen 1). Pipeline scaffolding from 32-agent research synthesis.