Havuz

qMLX: Custom inference engine for Qwen 3.5 122B on Apple Silicon, extending MLX with hybrid attention support, SSD-backed KV cache, and RYS

llama.cpp speculative decoding measured on one RTX 3090, Qwen3.6-35B-A3B UD-Q4_K_XL, commit 3737e4137. Published figures are re-derived from

UI-UX: a 4B multimodal LLM for mobile UX defect diagnosis, with UXBench — the first vision–language benchmark for UX reasoning (CVPR 2026 Fi

Any Model for VS Code Copilot

A powerful, lightweight agentic desktop agent built for engineers and work.

Pixel-Pets: A local, open-source virtual pet ecosystem on M5Stack. 100% AI-authored C++ (Claude) by a 10yo maker. Includes Muffin (LLM), Vis

AI coding agent in your terminal. Works with any LLM - DeepSeek, Qwen, GPT-5, Claude, Gemini, Kimi, Ollama. 100+ models via litellm. ~1300 l

Run Qwen Code as a coding agent through an OpenAI-compatible API.

Can't find subtitles? Generate SRT/ASS subtitles from any video with local AI and auto-translate them — no internet, free. macOS & Windows.

阿亮开源的 Skills 合集：短剧成片 / 电商详情图 / 儿童故事 / 有声绘本 / 配音 / 多人播客。一句话触发，AI 走完整 AIGC 流程。

让 Codex 白嫖 NVIDIA 免费顶级模型，一键接入 | Transparent proxy between Codex and NVIDIA NIM — use NVIDIA's free hosted top-tier models with Codex

A web-based memory usage and performance calculator for Huggingface GGUF models

一键配置各类 AI Agent 工具的 MCP、Skills 和规则；支持 Claude Code、Codex、DeepSeek Harness、OpenCode、Cursor、Gemini CLI、Qwen Code 等。 One-click local setup for M

A local-first desktop application for AI agent workflows.

Multimodal AI studio powered by Qwen3.6-35B-A3B. End-to-end web app exposing visual reasoning, image captioning, and document understanding

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

The open-source Three.js-based modern LLM Engine that runs in your browser via WebGPU compute.

OpenAI-compatible local AI proxy for DeepSeek, Claude , ChatGPT and Qwen — pipe any of them into VS Code or Terax using your real browser se

🦄 Forge - a local-first, multi-agent software-engineering runtime that runs Claude Code / Codex-style agentic workflows entirely on your own

An evidence-grounded LLM agent for diagnosing PyTorch training failures and verifying proposed repairs through bounded retraining.

Run a 177B model (Qwen3.8-Flash-Next) on an 8 GB laptop GPU. Measured: 6.6 GiB VRAM, 47.8 GiB RAM, 34-35 tok/s. The n-gram table stays on di

Your coding agent starts every session with amnesia — memo fixes that, 100% on your machine. Persistent memory for Claude Code, Codex, Curso

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

Locally fine-tuned Russian RAG models: document splitter, query expansion, retrieval embedder. Teacher distillation → GGUF → llama.cpp on AM

Free & Open Source AI Agent — Self-Hosted, Self-Learning, No Vendor API. Qwen2.5-7B + LoRA. 35 tools. Built on Islamic Epistemology (IHOS).

A local-LLM privacy layer for anything you paste into a cloud model — detect, tokenize, restore. Zero deps, reversible, strong on Chinese.

Evaan — a fully local, offline-capable AI companion chatbot built with Python & Hugging Face Transformers, running on Qwen2.5-0.5B-Instruct

A zero-allocation, header-only C++ BPE tokenizer for Qwen, built for maximum inference throughput.

A NPC AI Framework , based on character personalities, can be used for game NPC (extendable for Godot / Unity)

Qwen3.8-27B on one laptop CPU: up to 2.52 token/s, 8 GB tested, no runtime accuracy loss. Native OpenAI-compatible API with function tools;

The best model under 200B at near 10 token/s exact batch throughput on a single laptop CPU, with an automatic 8 GB RAM path. Native C, no GP

AI model that animates any SVG into Lottie with 3-4x fewer tokens than OmniLottie (CVPR 2026). Upload SVG → get animated .lottie. 3B model,

BrighTO LLM Router: free open-source, ultra-fast Rust LLM gateway/router and LiteLLM/Bifrost alternative with OpenAI + Anthropic APIs, Claud

食毅：已上线的 AI 饮食记录微信小程序，采用多模态识别、Human-in-the-loop 与确定性营养护栏

Beastmode: MofA (Mixture of Agents) orchestration framework for Hermes/OpenClaw/Codex with MemroOS-style context continuity.

On-device streaming speech-to-text for macOS, built on Qwen3-ASR via MLX

3D explorable atlas of every neuron in a transformer LLM. Click neurons, trace residual highways, compare cross-architecture conservation. L

Logic-based multi-LLM voice router for Home Assistant, combining native intents, local Qwen, GPU-aware desktop Ollama, and cloud AI with int

Companion code for the Manning book 'LLM Customization and Fine-Tuning.' Adapt open-weights LLMs end to end: prompting and RAG, LoRA/QLoRA,

Deploy Qwen3.6-35B-A3B (Q4_K_XL) + MTP speculative decoding on a single NVIDIA L4 24GB — GCP g2-standard-8 — via the official llama.cpp Dock

An autonomous agentic pipeline that finds, proves, and patches real C memory-safety vulnerabilities end-to-end using a single 7B open-source

Run SemIf (Jev-style semantic-if decisions) on a CPU — no GPU. Reads typed option probabilities straight from an open model in one forward p

Private, offline AI coding agent for local Qwen models via Ollama — zero dependencies, real tools, multi-agent. Your code never leaves your

👻 MoEWatch: A lightweight diagnostic and audit library for Mixture-of-Experts (MoE) models in HuggingFace Transformers — the pytest for MoE

生产级 LangGraph + FastAPI AI Agent 框架，支持阿里云 Qwen / DashScope，含中文注释、长期记忆、向量检索完整链路。

Open-source local AI server configs, GFX906 runtime maintenance, reproducible benchmarks, and QC methods for affordable AI research infrastr

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

中文公司调研 Agent

Agentic RAG consumer decision agent with ReAct, Text2SQL, multi-hop retrieval, memory and deterministic constraint checking.

Use third-party and Chinese model APIs as bounded Codex subagents while keeping Codex as the main agent.