xClean Tools

AI Repos — Daily Top 5

UPDATED 2026-11-03 01:12 PST

2026-10-13 · TUESDAY · 13:16 PDT

  1. 1 harry0703/MoneyPrinterTurbo +404 STARS TODAY · ★ 130,506 · PYTHON MoneyPrinterTurbo turns a topic or keyword into a finished vertical short video, writing the script, matching footage, adding subtitles and background music and compositing in HD. It runs from a web UI or an API on Windows, macOS and Linux, and can be driven by hosted models such as Kimi K3. The MIT-licensed Python project has about 130,000 stars and added roughly 400 today.
  2. 2 mvanhorn/last30days-skill +103 STARS TODAY · ★ 64,151 · PYTHON last30days is an agent skill that searches the past month of Reddit, Hacker News, Polymarket and GitHub in parallel, then has an agent judge synthesize one brief scored by upvotes, likes and prediction-market money rather than editorial picks. A setup wizard adds X, YouTube, TikTok and arXiv. It installs into Claude Code or any skills host; about 64,000 stars, 103 added today.
  3. 3 wuyoscar/AISafetyHot-Hub ★ 1,237 · TEX AISafetyHot-Hub installs six agent skills plus an MCP server that pull AI safety material - jailbreaks, prompt injection, agent security, evaluations, incidents and policy - into Claude Code, Codex or Cursor. The skills cover 24-hour and 7-day briefings, paper scouting that returns BibTeX and related-work drafting, and the project posts about 30 selected papers a day. No API key needed.
  4. 4 jwuthri-cloud/Tracely-ai ★ 738 · PYTHON Tracely is trace-native CI for AI agents: it grades production agent traces as they land, clusters the failures into issues, freezes bad runs into hermetic replayable cases and blocks the pull request that would ship the same failure again. The whole stack - API, worker, UI, Postgres, ClickHouse, Redis and MinIO - self-hosts in one click, with a Python SDK on PyPI. MIT licensed.
  5. 5 Jakeschincariol/founder-skill ★ 536 · PYTHON founder-skill is a set of eleven Claude skills that stress-test a business before launch, running as subagents in Claude Code. They include an advisory board the author says is trained on Hormozi, Thiel and Jobs, a marketing director, a CFO and a consumer panel of 100 buyer agents, with unit economics and business plans covered. MIT licensed, free.

2026-10-12 · MONDAY · 13:14 PDT

  1. 1 OpenBMB/VoxCPM +370 STARS TODAY · ★ 39,162 · PYTHON The VoxCPM repository now ships VoxCPM2, a tokenizer-free text-to-speech system that generates continuous speech representations end to end instead of discrete audio tokens. It covers multilingual synthesis, prompt-driven voice design and voice cloning. The Python project added about 370 stars today for a total near 39,200.
  2. 2 multimodal-art-projection/YuE +318 STARS TODAY · ★ 11,788 · PYTHON The YuE repository keeps trending on its YuE2 release, adding about 318 stars today for a total near 11,800. Built by HKUST, M-A-P and collaborators, YuE2 unifies symbolic and audio music generation, performs zero-shot covers, and supports agentic editing of existing tracks. The 3B weights, a technical report and a song benchmark are public.
  3. 3 open-webui/open-webui +142 STARS TODAY · ★ 154,817 · PYTHON Open WebUI, the self-hosted web front end for chatting with language models, added about 142 stars today and now stands near 154,800. It works with Ollama and any OpenAI-compatible API, and bundles retrieval over user documents plus MCP tool support. It remains one of the most-starred AI projects on GitHub.
  4. 4 microsoft/BitNet +134 STARS TODAY · ★ 40,843 · C++ Microsoft's bitnet.cpp, the official inference framework for 1-bit LLMs, added about 134 stars today to reach roughly 40,800. Its kernels run 1.58-bit models losslessly on CPU and GPU, with reported speedups of 1.37x to 5.07x on ARM and 2.37x to 6.17x on x86 against F16. It can serve a 100B BitNet b1.58 model from a single CPU at 5 to 7 tokens per second.
  5. 5 deepseek-ai/DeepGEMM-Ascend ★ 573 · C++ DeepSeek has published DeepGEMM-Ascend, a port of its DeepGEMM matrix-multiplication library to Huawei Ascend NPUs. It keeps the same API and covers BF16, FP8 and FP4 GEMM, MQA logits and MegaMoE, wrapping Ascend matrix multiply-add primitives so kernels avoid fractal layout and alignment detail. The initial release on September 30 targets Ascend 950 parts.

2026-10-11 · SUNDAY · 13:09 PDT

  1. 1 lyogavin/airllm +541 STARS TODAY · ★ 36,331 · JUPYTER NOTEBOOK AirLLM added about 541 stars on the day. The Apache-2.0 library streams a model's weights layer by layer so large models run on small cards without quantization, distillation or pruning, and its own measurements put Kimi K3 at 2.8T parameters under 4 GB of VRAM and Qwen3.8-Flash-Next at 125B in about 6 GB. A September release extended the same trick to training.
  2. 2 Alishahryar1/free-claude-code +461 STARS TODAY · ★ 57,977 · PYTHON free-claude-code added about 461 stars on the day. The Python project routes Claude Code, Codex, VS Code, Pi, OpenCode and six other harnesses to free token allowances it advertises at more than 1.3 billion, usable from a terminal, app, IDE or phone. It has since added native browser sessions with multi-harness and multi-model support.
  3. 3 repowise-dev/repowise +68 STARS TODAY · ★ 7,520 · PYTHON Repowise added about 68 stars on the day. The AGPL-3.0 tool indexes a repository's call graph, git history, tests and docs on the user's own machine, then answers what calls what, what breaks on a change and what is dead, serving the index to coding agents over MCP. Its graph, risk, health and dead-code passes make no LLM calls and need no API key.
  4. 4 rociiu/talorys ★ 728 · TYPESCRIPT Talorys is an MIT-licensed personal AI assistant that runs entirely inside the user's own Cloudflare account, installed with one npx command. Chat, memory, tasks, notes and reminders live in a Durable Object with SQLite, scheduled on alarms so nothing has to stay online, and the agent Worker ships with no public URL. It keeps working without AI if the Workers AI quota runs out.
  5. 5 sirioberati/Genjustsu-Open-Source-Workflow ★ 612 · PYTHON Genjustsu is an MIT-licensed local workflow from TEIN for replacing the subject of a video. It separates the audio with Demucs, normalizes the clip, derives colored depth, cuts SAM 3 masks and sends the composite to Seedance through Enhancor, gating every billed generation behind a full-clip mask review and a separate 1080p approval. The export restores the entire original soundtrack rather than generated audio.

2026-10-10 · SATURDAY · 13:13 PDT

  1. 1 multica-ai/andrej-karpathy-skills +279 STARS TODAY · ★ 218,128 The repo packages a single CLAUDE.md, with Cursor and Claude plugin variants, holding four rules for coding agents: think before coding, prefer the simpler design, keep edits surgical, and work toward verifiable goals. It is drawn from Andrej Karpathy's public notes on how models make silent assumptions and bloat code. The project added about 279 stars on the day.
  2. 2 huggingface/transformers +94 STARS TODAY · ★ 167,161 · PYTHON Transformers is the model-definition framework behind much of the open-weights ecosystem, covering text, vision, audio and multimodal architectures for both inference and training. The Apache-2.0 Python library holds the reference implementations that hub checkpoints load from, and is a common dependency of fine-tuning and serving stacks. It added about 94 stars on the day.
  3. 3 blendi-remade/agentcraft ★ 677 · JAVA AgentCraft runs a team of Claude Agent SDK workers inside a Minecraft studio: a lead agent reads the repo and pins tasks to a wall, while workers code in their own git worktrees and in-game monitors stream the files they read and the lines they change. A merge lands only when the player reviews the real diff and clicks, and nothing is pushed. The Fabric mod keeps agents working while the game is closed.
  4. 4 GanyuanRan/Autoloom ★ 671 · JAVASCRIPT Autoloom is a free desktop client for coding agents that builds its Aegis governance methods into the run loop: it asks whether a change is needed at all, flags added structural complexity, requires evidence-grounded diagnosis after a failure, and reviews delivery against the stated requirements. Built on DeepSeek's open-source Harness, it ships as a Windows x64 beta and works with the user's own choice of model.
  5. 5 DozenTwelve/Papermorph ★ 658 · HTML Papermorph is an agent skill that turns a PDF into an animated, narrated web book, running a pipeline of book plan, storyboards, narration, then animations and quizzes. Built with Opus 5.5 and tested on other models, it ships two math titles on a public bookshelf, with physics chapters still being refined. Image models, multilingual output and background music are not supported yet.

2026-10-09 · FRIDAY · 01:13 PDT

  1. 1 Edwardxlai/easyread ★ 871 · PYTHON EasyRead is an MIT-licensed desktop app that turns English research papers into typeset Chinese. It translates PDFs page by page while keeping formulas in KaTeX and tables laid out as in the original, and adds side-by-side comparison with the source page, highlighting, notes and an ask-the-AI panel. Translation runs through Claude Code, Codex CLI or API keys, and the library stays on the user's own machine.
  2. 2 LosaLosSantos/aurelio-finance ★ 701 · PYTHON Aurelio is a self-hosted wealth-management app built on FastAPI and React under an MIT license. It tracks accounts at banks and brokers, positions priced from the market with cost, gain and dividends, net worth over time, and fund look-through by country, sector and company. Records stay in a local database file, and the AI advisor is optional, going through OpenRouter with the user's own key.
  3. 3 elstongun/leviathan ★ 698 · RUST Leviathan is a single-binary Rust tool that turns JSONL, CSV, SQLite and other exported records into a ranked full-text index that agents can query in plain words. It answers with short cited result cards; on a synthetic one-million-record maintenance log, its benchmark reports a median of 436 tokens per question against about 107,000 for the best grep strategy. Apache-2.0 licensed, with optional MCP support.
  4. 4 AgentSystemLabs/agent-office ★ 687 · TYPESCRIPT Agent Office seats coding agents at desks in a shared 3D office, with each GitHub repo as a floor of the building. Teammates can watch any worker's terminal, drop into it together, talk over voice and track issues and pull requests; it supports Claude Code, Codex, OpenCode, Grok, Muse, DeepSeek Harness and Cursor. The MIT-licensed project warns it is a fast-moving work in progress.
  5. 5 AgentMemoryRepo/agentmemoryrepo ★ 655 Agent Memory Repo is a spec for giving coding agents memory that persists across sessions, saved in real time as linked entries the agent navigates like a wiki. The memories live in a separate private Git repository rather than the public spec repo. Cognition publishes the spec on its site and ships it as an installable skill for Devin CLI, Claude Code, Cursor and other agents.

2026-10-08 · THURSDAY · 13:15 PDT

  1. 1 CursorTouch/Windows-MCP +411 STARS TODAY · ★ 8,308 · PYTHON Windows-MCP, an MIT-licensed MCP server that lets LLM agents drive the Windows desktop, added about 411 stars today. It exposes keyboard and mouse control, window and UI state capture, and a DOM mode for browser automation, and works with any model without computer vision or fine-tuning. It supports Windows 7 through 11.
  2. 2 microsoft/agent-framework +24 STARS TODAY · ★ 14,018 · PYTHON Microsoft Agent Framework is an open multi-language framework for production AI agents and multi-agent workflows, with SDKs for Python, .NET and Go. It provides graph-based orchestration patterns including sequential, concurrent, handoff and group collaboration, plus middleware, durability and human-in-the-loop control. It connects to Microsoft Foundry, Azure OpenAI, OpenAI and the GitHub Copilot SDK.
  3. 3 lemomo-ai/lemo-opuscar ★ 1,410 · JAVASCRIPT Lemo-Opuscar is a Claude Code skill for making short films without using a video model. It ships 43 film styles, each pairing a style prompt with a demo film rendered entirely in code, alongside directing and technical guides. The demo films were produced with Claude Opus 5.5.
  4. 4 oil-oil/oil-ui ★ 1,083 · PYTHON Oil-UI is an agent skill for pushing AI user-interface design further by exploring several distinct directions at once. It lays the candidate designs out side by side for comparison and refines them against real screenshots. The project targets Claude Code and Codex.
  5. 5 echris6/motion-video-kit ★ 1,069 · PYTHON Motion-Video-Kit is a Claude Code skill kit for AI-assisted business videos. It bundles an independent critic loop, motion principles drawn from 28 launch films, a quality bar, sound design guidance, Three.js patterns and helper scripts.

2026-10-07 · WEDNESDAY · 13:14 PDT

  1. 1 manaflow-ai/cmux +96 STARS TODAY · ★ 27,797 · SWIFT cmux is an open-source macOS terminal built on Ghostty and aimed at running several AI coding agents side by side. It adds vertical tabs, split panes and per-session notifications, and its sidebar tracks each session's git branch, linked PR status, working directory and listening ports. A scriptable browser pane and SSH workspaces sit alongside the terminal.
  2. 2 vllm-project/vllm +73 STARS TODAY · ★ 93,338 · PYTHON vLLM is an open-source inference and serving engine for large language models, built for high throughput and efficient memory use. Its PagedAttention scheme manages the key-value cache in fixed-size blocks and continuous batching keeps accelerators busy, and it ships an OpenAI-compatible API server across a range of GPU and accelerator backends.
  3. 3 MDX-Tom/gpt-instruct +43 STARS TODAY · ★ 9,315 · PYTHON gpt-instruct is an MIT-licensed prompt and evaluation pack aimed at getting past the safety training of Codex-style GPT coding models. The repository ships versioned prompt archives for three named GPT releases alongside a Python runner, unit tests and a JailbreakBench dependency list for scoring results. It has gathered more than 9,000 stars since July.
  4. 4 uber/ADR +36 STARS TODAY · ★ 1,891 · PYTHON Uber open-sourced ADR, or Agentic AI Detection and Response, the security system it runs in production to watch AI agents. It inventories AI tools on employee endpoints and captures agent intent and tool-use traces across more than seven coding agents on macOS, Linux and Windows, and ships ADR-Bench with over 300 tasks and 134 MCP servers. The paper was accepted to MLSys 2026.
  5. 5 allenai/olmocr +22 STARS TODAY · ★ 19,723 · PYTHON olmOCR is Ai2's toolkit for turning PDFs and page images into clean Markdown for LLM training data. It handles equations, tables, handwriting and multi-column layouts, strips headers and footers, and recovers a natural reading order. It runs on a 7B vision-language model, so it needs a GPU, and Ai2 puts the cost under 200 dollars per million pages.