Havuz

OAK, C++ ile cihaz üzerinde AI agent çalıştırmayı hedefleyen, bazı agent özelliklerini açık kaynak modüller olarak sunan alpha aşamasındaki bir projedir.

Neden yükseliyor · 48 günde 1170 yıldız almış; on-device AI, gizlilik ve Qualcomm NPU desteği ilgi çekici bir konumlandırma. Ancak yükselişin özel nedenini mevcut veriden belirlemek mümkün değil. Son 28 günde commit ve farklı yazar sayısı sıfır; README’deki performans ve karşılaştırma iddiaları da burada bağımsız doğrulanmış kanıt sunmuyor.

audio.cpp, ggml tabanlı ortak bir C++ runtime üzerinden farklı ses modellerinde yerel inference çalıştırmayı sağlayan bir framework’tür.

Neden yükseliyor · 96 günlük yaşına rağmen 3067 yıldızın yanı sıra son 28 günde 272 commit ve 18 farklı yazar görünüyor; bu, README’deki geniş model desteğinin yanında kayda değer bir geliştirme hareketliliğine işaret ediyor. README performans iyileştirmeleri ve yeni model desteği bildiriyor, ancak bu kıyasların bağımsız doğrulaması verilen bilgilerden anlaşılamıyor.

Needle, küçük cihazlarda tool calling, structured extraction ve embedding çalıştırmayı hedefleyen kompakt bir model ve yazılım paketidir.

Neden yükseliyor · Küçük cihazlarda yerel tool calling ve extraction ihtiyacına odaklanması, 216 günde 12.730 yıldıza ulaşması ve son 28 gündeki 36 commit ile 15 farklı yazarın katkısı dikkat çekiyor. Yıldızlar ilginin göstergesi; model performansının bağımsız doğrulaması değil.

On-device, ComfyUI-style node graph for image/video diffusion on Android. Run MiniMax H3 and WAN 2.2 14B on the Qualcomm Hexagon NPU. SD1.5,

Run MoE models bigger than your RAM. Frontier-size MoE on a phone, laptop, edge. Lossless, on stock llama.cpp. An adaptive streaming optimiz

Whistle speech-to-text running fully on an ESP32-S3: hold a button, say a colour in one of 7 languages, the LED does it. No network.

Offline AI research assistant for Android. Answers from Wikipedia and other reference packs on the phone, with citations. Gemma 4 runs on th

On-device AI SDKs for iOS, macOS, Android, and the web. Small, focused models that run fully offline in Swift, Kotlin, and JavaScript with C

Open Source Brain Platform | AI training and inference in Rust

Ultralytics YOLO26 quickstart for detection, instance and semantic segmentation, depth estimation, classification, pose, OBB, and tracking.

Oído, ESP32-S3 üzerinde cloud veya NPU olmadan open-vocabulary speech recognition çalıştırmayı hedefleyen C tabanlı bir ASR motorudur.

Neden yükseliyor · Düşük maliyetli bir mikrodenetleyicide open-vocabulary ASR çalıştırma iddiası ve README’de verilen LibriSpeech sonuçları dikkat çekiyor. Repo 1 günlük ve 46 yıldız almış olsa da son 28 gündeki 6 commit’in tek yazardan gelmesi, ilginin henüz bağımsız benimseme veya sürdürülebilir geliştirme kanıtı olmadığını gösteriyor.

Ultralytics YOLOv8 discovery and quickstart for detection, segmentation, classification, pose, OBB, and tracking.

Optimized ESP32-S3/P4 fully offline speech to text and text to speech system for Micropython and AtomVM, featuring a finetuned STT model for

Private on-device AI chat for Android — runs any GGUF model locally via llama.cpp with ARM-optimised SIMD. Zero network permissions, encry

A multi-model Muse: personal AI voice assistant on ESP32 - wake word or knob, any OpenAI-compatible brain (STT/chat/TTS mix-and-match), stre

🔥 Run large open language models on your phone. A C++20 GGUF inference engine that streams models larger than RAM from flash, paces itself b

Ultralytics YOLO11 discovery and quickstart for detection, segmentation, classification, pose, OBB, and tracking.

Natural-sounding streaming text-to-speech for the ESP32-S3: 3.3 M params, 3.8 MB of weights, no cloud, no NPU

On-device simultaneous interpretation on Apple Silicon — VoxCPM2 on MLX at the core, streaming TTS in ~170 ms, voice clone, any-pair transla

Your car as a chat-room agent: Raspberry Pi 5 + dashcam + local AI. CodeWatch's sibling for the garage.

Fastest MoE/LLM inference runtime for consumer and edge Blackwell GPUs. SN74 on Gittensor.

Deploy streaming ASR + TTS on RK3576/RK3588 — 120ms TTS, 52-language ASR, one Docker command

Proof of concept: turn old Android phones into an LLM inference cluster. ADB zero-touch provisioning, llama.cpp on each phone, OpenAI-compat

A ternary, zero-heap tiny language model that runs inside a $2 microcontroller — bit-exact Python <-> C99 <-> Cortex-M3 (QEMU) parity. Apach

The 30 MB open-source edge AI agent runtime. Run AI agents offline on Linux (Raspberry Pi, Jetson). GPIO, UART, MQTT as first-class nodes. I

600 KB WASM runtime for Cactus Compute's Needle AI tool-calling models, Needle 3, 2 and 1 from one build. Browser, Cloudflare Workers, Node.

Local-first AI memory — runs offline on any machine with 8 GB+ RAM (SBC, mini PC, laptop, workstation). Zero-loss verbatim archive, knowledg

Offline speech-to-text on the STM32N6 microcontroller: free English speech recognition on the NPU, no cloud

Low-memory C++20 LLM inference runtime for quantized GGUF models on CPU — mmap-backed weights, paged KV cache, Q4/Q8 kernels, and reproducib

[Archived] AAOSP, predecessor to JibarOS. Early experiment in OS-level on-device AI (MCP + local LLM on Android 15). See JibarOS for the act

面向智能座舱的云边协同 AI Agent：端侧毫秒级车控，云端声明式 Multi-Agent 与 Skill/DAG 编排，支持 S2S 实时语音、声纹多用户、HMI和Android双端；LLM 只负责理解与规划，VAL 负责确定性安全执行。

Drop-in ggml backend for Rockchip NPUs: offloads llama.cpp/whisper.cpp prefill to the RK3588 NPU

P2P semantic matching for on-device AI agents. No server, no cloud.

I Ran Qwen 3.8 27B on Just 4GB VRAM! (FreeToken Tutorial) - Edge-native MoE serving engine for local LLMs on low VRAM GPUs

Tiny, Apache-2.0 wake-word detection you train yourself and run on-device in Python, the browser, and on mobile.

CPU-first LLM inference server on llama.cpp. Runs useful models on free-tier ARM boxes; rewriting the input made it ~6x faster and more accu

Distributed 56M-parameter LLM inference across 3 ESP32-S3 boards via ESP-NOW , Split-PLE + KV cache, fully offline.

Fully local voice AI for iOS

FastCrest Tether: the OSS edge-to-cloud AI deploy CLI. Optimize, verify, deploy across Jetson, RTX, Apple Silicon, AMD. Hybrid edge-cloud in

GLM-5.2, a 744 billion parameter mixture of experts model, in a pure C inference engine: quantized to int4, experts streamed from disk, depl

Run a 28.9M-parameter TinyLM on ESP32-S3 with an RP2040 OLED display node for fully local embedded AI inference.

edge-dit.cpp — a native C/C++ inference engine for Diffusion Transformers (DiT), designed for local and resource-constrained devices with au

CPU-side action routing, context compression, and causal memory for AI agents — matches an LLM-everything agent at 58% fewer LLM calls and ~

Lightweight AI agent runtime for homelabs, built around deterministic skills and local LLMs.

[MM' 2026] Light-MER for efficient multimodal emotion recognition with sub-1B MLLMs.

Local inference you can prove. A single-binary GGUF engine in C: serves, scores, and trains LoRA through the quantized weights it deploys, b

Distributed, peer-to-peer, decentralized network for LLM inference — peers seed & leech completions, metered by BitTorrent-protocol-style up

🧠 One LLM, split across a Mac (Apple MPS) + a Windows PC (NVIDIA CUDA) — heterogeneous pipeline inference over a framework-neutral wire, bit

LLM-free, layout-aware PDF chunking for RAG pipelines. Preserves document structure via regex and font heuristics. Written in pure Rust.