LLMs & Model Tooling
Model-layer infrastructure: training, fine-tuning, prompting and everything that wraps a large language model.
Tamper-evident runtime records for AI agents. Hash-chained, dependency-free, verifiable by anyone.
Apple Silicon mlx with Zero Dependency Java
A tiny MCP stdio server that exposes DeepSeek as a cheap supervised worker for Claude Code, Codex, or any MCP client.
The AI-native JupyterLab — open-source Cursor for notebooks. Cmd+K inline edit, multi-step agent with cell-level tools (read/edit/run), chat with @cell/@file context, ghost-text completion, one-click traceback fix. BYO model: Anthropic, OpenAI, Gemini, Ollama, vLLM. Local-first, privacy-first, fully open source.
Open-source metasearch backend, MCP server, and AI search API for LLM agents. Python FastAPI search gateway with Google search via SerpBase and Serper, multi-engine search aggregation, structured JSON output, provider fallback, deduplication, and SearXNG alternative architecture for agent workflows.
餐饮连锁数字化 + AI 全景蓝图:原串(平价烤串连锁)的架构决策、业务口径、踩坑实录与可直接喂给 AI 的复刻指令。纯自然语言方案,不含代码与任何真实经营数据。
Research-backed agent skills and tools for premium image, video, audio, voice, and generative media production across AI coding assistants.
English text humanizer for Claude: typography, vocabulary, structure. Calibrates to your voice. Built on the UMD / Google DeepMind study and Wikipedia's Signs of AI writing
One Agent. One memory. Shared by every AI. 🍙
A curated collection of 51 production-ready subagents for Gemini CLI. Drop them into .gemini/agents/ and let Gemini delegate the right specialist.
Open-source AI coding agent for your terminal — one Rust binary, 15+ LLM providers, 1,000+ models. A self-hostable Claude Code / Codex CLI alternative (MIT).
Self-hosted web viewer for Claude Code session transcripts — read, search, replay, and audit every session you have ever run
Distil your genome down to what matters, and share only that. Local-first longevity profiles from consumer DNA files, for discussing with an LLM without handing over all 640,000 variants.
Observable diagnostics, failure attribution, metrics, and no-key replay for multi-agent runtimes.
SSD-streaming inference engine for giant MoE models (Rust + CUDA). GLM 5.2 743B at 2 tok/s and Hy3 295B at 7 tok/s on two consumer 16GB GPUs. Zero-config multi-GPU: measures PCIe bandwidth, places attention and hot experts where they fit.
Build & prompt Qwen3.x agents in their native dialect — extracted from qwen-code & Qwen-Agent
Evidence-grounded market research prototype with traceable AI workflows.
Multi-session IDALib MCP router for coding agents. Analyze multiple binaries in parallel with IDA-compatible reverse engineering tools.
Reusable AI prompt and project brief system for cinematic scroll-driven 2.5D websites.
Open-source marketing skills that run on any model — with a hosted brand-memory engine. No coding agent required.
A from-scratch implementation of the language-model training pipeline in PyTorch: tokenization, pretraining, SFT, and preference-based alignment.
Quality-per-token GPT-5.6 Sol, Terra, and Luna model router for Codex, Work, CLI, and subagents.
Embed the WorkBuddy/CodeBuddy CLI as an AI chat agent inside Obsidian
Architect-first delegation for Claude Code and OpenCode. A Fable/Opus session routes implementation to Codex, the OpenCode provider pool, local Pi ($0), or autonomous Pythinker, then reviews every diff.
Architect-first delegation for Claude Code and OpenCode. A Fable/Opus session routes implementation to Codex, the OpenCode provider pool, local Pi ($0), or autonomous Pythinker, then reviews every diff.
Offline prompt engineering manager for Windows
Anti-AI-slop agent skill: makes text read written and UI look made, not generated. Deterministic scanners + severity-tiered tell catalogs.
把整个浏览器交给大模型 | Hand the whole browser to an LLM — 本地开源的浏览器 Agent + 网页逆向工具。observer 架构只把浏览器最新状态暴露给模型:既能点击抓取,又能下 CDP 断点、逆向签名/token/加密、生成脱离浏览器的爬虫。A local browser agent & web reverse-engineering tool with a desktop UI.