GitHub Trends

#jupyter_notebook

NVIDIA Cosmos is an open platform with world models, datasets, and tools for building Physical AI for robots, self-driving cars, and smart systems. Cosmos 3 can understand text, images, and video, and can also generate images, video, sound, and action sequences. This helps you create smarter AI that can plan, predict, simulate, and train faster, with ready paths for research and production.

https://github.com/NVIDIA/cosmos

#javascript #ai #anthropic #caveman #claude #claude_code #llm #meme #prompt_engineering #skill #tokens

This tool helps agents answer in caveman style: fewer words, same technical meaning, and often much lower token cost. You can choose levels like lite, full, ultra, or wenyan, and it also includes commands for shorter commit messages, quick PR reviews, stats, and compressing memory files; the benefit is faster replies, lower cost, and longer context for your work.

https://github.com/JuliusBrussee/caveman

#python #ai_agent #algorithmic_trading #backtesting #fintech #llm #mcp #multi_agent #python #quantitative_finance #trading

Vibe-Trading is an open-source trading research tool that turns plain English questions into market analysis, backtests, reports, and trading exports. It helps you test ideas, study your own trades, and run agent teams, so you can make faster, better trading decisions without building everything from scratch.

https://github.com/HKUDS/Vibe-Trading

#python #agent #ai #anthropic #claude_code #compression #context_engineering #context_window #cursor #fastapi #langchain #llm #mcp #openai #prompt_engineering #proxy #python #rag #token_optimization #tokens #typescript

Headroom is a local tool for AI agents that shrinks prompts, logs, files, and chat history before sending them to an LLM, often cutting tokens by 60–95% while keeping the same answer quality. It can work as a library, proxy, MCP server, or agent wrapper, so you can save tokens, speed up workflows, and still recover the original content when needed.

https://github.com/chopratejas/headroom

#python #ai #ai_companion #ai_vtuber #ai_waifu #chatbots #live2d #live2d_web #llm #neuro_sama #ollama

Open-LLM-VTuber is an offline AI companion with voice chat, vision, and a Live2D avatar that works on Windows, macOS, and Linux. It can run as a web app or desktop pet, supports many LLM, speech-to-text, and text-to-speech options, and is highly customizable, so you can build a personal assistant that fits your style and keep chats stored for later.

https://github.com/Open-LLM-VTuber/Open-LLM-VTuber

#other #rime #rime_schema

This list shows many Rime input methods and setups for Chinese, dialects, and other languages, including pinyin, double pinyin, Wubi, Cantonese, Japanese, Korean, and emoji. The benefit to you is that you can quickly find a scheme or configuration that matches your typing style, language, or device, which can make typing faster and easier.

https://github.com/ayaka14732/awesome-rime

#javascript #agent_orchestration #agent_skills #agentic_ai #agentic_workflow #ai_agents #ai_automation #babysitter #claude_code #claude_code_skills #claude_code_workflows #claude_skills #claude_workflows #codex_plugin #codex_skills #codex_workflow #hermes_plugin #pi_extension #trustworthy_ai #vibe_coding

Babysitter is a tool that helps you run AI coding work in a strict, step-by-step way, with checks, approvals, and a full log of every decision. It can save you time, reduce mistakes, and make complex tasks easier to control because it keeps the process deterministic, resumable, and easier to audit.

https://github.com/a5c-ai/babysitter

#python #agent #ai_agents #hermes #hermes_agent #nous_research

Hermes WebUI gives you a simple browser interface for the Hermes Agent with the same main power as the command line, plus chat, file browsing, session tools, and settings in one place. It keeps memory across sessions, can run jobs while you are away, and is easy to start with one command. The benefit to you is faster work with less retyping, because it remembers your setup, lets you use it from a browser or phone, and keeps your conversations and files on your own server.

https://github.com/nesquena/hermes-webui

#typescript

Pi Subagents helps you split work into focused child agents for review, planning, research, coding, and background jobs, with just one install command and simple plain-English prompts. This saves you time and gives you extra checks, so you can catch mistakes, compare ideas, and finish complex work more safely.

https://github.com/nicobailon/pi-subagents

#python #bilibili #douyin #tiktok #xiaohongshu #youtube

This tool helps you upload videos and schedule posts to major platforms like Douyin, Bilibili, Xiaohongshu, Kuaishou, WeChat Channels, Baijiahao, and TikTok from one place. It saves you time by automating repetitive posting work, so you can manage more accounts with less manual effort and focus more on creating content.

https://github.com/dreammis/social-auto-upload

#dart #clash #clash_meta #flutter #hysteria #multi_platform #proxy #v2ray #vless #vpn

FlClash is a free, open-source proxy app for Android, Windows, macOS, and Linux that is simple to use, has no ads, supports dark mode, themes, and WebDAV sync, and works with subscription links. For you, that means one app can help manage proxy settings across devices with a clean interface and easy setup.

https://github.com/chen08209/FlClash

#rust #document_ocr #document_processing #ocr #ocr_recognition #pdf #pdf_parser #text_extraction

LiteParse is a fast, local PDF parser that extracts text with bounding boxes, can use OCR, and works in Rust, Python, Node.js, and the browser. It also makes screenshots and can handle files like DOCX, XLSX, PPTX, and images after conversion. Benefit: you can turn documents into clean text or JSON on your own machine, which helps with private, quick, and structured document processing.

https://github.com/run-llama/liteparse

#typescript

Cursor plugins are ready-made add-ons for Cursor, each kept in its own folder with its own manifest file. The repository also has a master list of all plugins, plus files like skills, rules, MCP settings, README, changelog, and license, which helps you find, manage, and build plugins faster and more clearly.

https://github.com/cursor/plugins

#python #deep_learning #jepa #model_predictive_control #pytorch #world_model

stable-worldmodel is a tool for world model research that lets you collect data, train models, and test them in one place across many standard environments. It helps you work faster and compare results more easily by giving you ready-made baselines, solvers, and simple data formats, so you can focus on improving your model instead of building the whole setup yourself.

https://github.com/galilai-group/stable-worldmodel

#html #claude_code #claude_code_plugin #harness #harness_engineering

Harness is a Claude Code plugin that builds agent teams and skills for your project from one simple prompt, using six team patterns like pipeline, supervisor, and fan-out/fan-in. It helps you turn a domain idea into organized agents, better coordination, and tested outputs, so you can save time and handle complex work more reliably.

https://github.com/revfactory/harness

#python #audio #audio_tokenizer #llm #multimodal #text_to_speech #voice_cloning

MOSS-TTS is an open-source family of speech and sound models for natural, high-quality audio generation, including voice cloning, multi-speaker dialogue, real-time speech, and sound effects. It supports 31 languages in v1.5, better voice stability, and pause control, and it also offers a lightweight Nano version that can run on 4 CPU cores. The benefit to you is simple: you can create realistic speech or sound for apps, demos, or products with strong quality, flexible control, and multiple ways to run it.

https://github.com/OpenMOSS/MOSS-TTS

#shell

Claude Code Harness is a tool that adds a clear loop for coding: plan, work, review, sync, and release. It helps you avoid messy agent work by writing a spec and plan first, checking the work, and saving proof for pull requests or releases, so you can ship safer changes with less rework and more confidence.

https://github.com/Chachamaru127/claude-code-harness

#other #chinese #english_learning #tutorial

This is an English learning guide that shares the writer’s methods, reading, listening, speaking, writing, vocabulary, and AI tips. Its benefit is simple: it helps you learn English more efficiently, find the right study path, and use AI as a real study tool instead of just a translator.

https://github.com/byoungd/English-level-up-tips

#typescript #coderabbit #inngest #nextjs #shadcn_ui #stock_market #tailwindcss

OpenStock is a free, open-source stock market app that lets you track prices, set alerts, search stocks fast, and view company details and charts. It uses Next.js, MongoDB, Finnhub, and TradingView, and it also includes watchlists, email updates, and personalized onboarding. Benefit: you get a modern market tool without paying for expensive platforms, and you can even run or modify it yourself because it is open source.

https://github.com/Open-Dev-Society/OpenStock

#python #cloth #collision #contact #physics #simulation

ZOZO’s Contact Solver is a GPU-based tool for physics simulations of cloth, solids, and rods that aims to prevent object overlap and handle very large scenes. It works with Blender, JupyterLab, Docker, Windows, Linux, and cloud servers, so you can run simulations locally or remotely and even use a laptop or Mac while the heavy work runs on a GPU. The benefit to you is faster setup, flexible use, and reliable, penetration-free results for animation, design, or testing.

https://github.com/st-tech/ppf-contact-solver