xClean Tools

AI Repos — Daily Top 5

UPDATED 2026-11-03 01:12 PST

2026-10-06 · TUESDAY · 13:12 PDT

  1. 1 deepseek-ai/DeepGEMM +363 STARS TODAY · ★ 8,646 · CUDA DeepSeek's DeepGEMM is a BLAS kernel library for matrix multiplication on GPUs, written in CUDA. The repository added 363 stars today and now holds 8,646. Kernel libraries of this kind set the practical throughput ceiling for training and inference on a given GPU.
  2. 2 cathrynlavery/diagram-design +227 STARS TODAY · ★ 43,916 · HTML diagram-design is a set of editorial diagram styles for coding agents including Claude Code, Codex, GitHub Copilot, Factory Droid and Pi. It ships 42 diagram types as self-contained HTML and SVG rather than Mermaid markup. The repository added 227 stars today for a total of 43,916.
  3. 3 anthropics/knowledge-work-plugins +124 STARS TODAY · ★ 26,339 · PYTHON Anthropic has open-sourced 11 plugins that turn Claude into a specialist for a given role or team, built for Claude Cowork and also compatible with Claude Code. Each plugin bundles the skills, connectors, slash commands and sub-agents for one job function. The repository added 124 stars today for a total of 26,339.
  4. 4 omnigent-ai/omnigent +50 STARS TODAY · ★ 10,622 · PYTHON Omnigent is an open-source agent framework and meta-harness that orchestrates Claude Code, Codex, Cursor, Pi and custom agents behind one interface. It lets users swap harnesses without rewriting, enforces policies and sandboxing, and supports real-time collaboration from any device. The Python project added 50 stars today for a total of 10,622.
  5. 5 raullenchai/Rapid-MLX +16 STARS TODAY · ★ 3,915 · PYTHON Rapid-MLX is an Apache 2.0 inference server for Apple Silicon built on MLX, exposing OpenAI- and Anthropic-compatible APIs. It is tuned for reliable tool calling by coding agents and claims up to 4x the speed of Apple's own mlx-lm on the same weights. The repository added 16 stars today for a total of 3,915.

2026-10-05 · MONDAY · 13:12 PDT

  1. 1 msitarzewski/agency-agents +595 STARS TODAY · ★ 157,162 · SHELL A roster of AI agent personas for coding harnesses, each given a defined voice, process and deliverables rather than a generic prompt template. A companion desktop app for macOS, Linux and Windows browses the whole collection and installs agents into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen and Osaurus, then keeps them updated.
  2. 2 VictorTaelin/OptMem +106 STARS TODAY · ★ 1,930 · PYTHON Permanent memory for coding agents: a 426-token prompt plus a single-file Python tool the agent calls to read, record and search its own memories. Each memory is one line in an append-only log, and the tree of summaries above it is a cache that can be rebuilt from the log alone. Records are fixed width, so a lookup is one seek and reading memory takes 0.03 seconds at a million entries.
  3. 3 shiyu-coder/Kronos +86 STARS TODAY · ★ 40,020 · PYTHON Kronos is a family of decoder-only foundation models pre-trained on financial candlestick sequences from more than 45 global exchanges. The authors present it as the first open-source foundation model for that data, with released weights on Hugging Face, a live demo and fine-tuning scripts. The paper was accepted at AAAI 2026.
  4. 4 omlahore/RemoveMacAI ★ 1,777 · SWIFT A command line tool that turns Apple Intelligence off on macOS 27, removes the downloaded models and stops macOS from fetching them again. macOS 27 no longer offers a single switch for the feature and leaves the models on disk once it is disabled. The tool reports each feature and its model size on disk, and every change can be reverted.
  5. 5 nykooi1/vibe-wise ★ 1,624 · PYTHON A Claude Code plugin that has the model ask for your approach before it writes anything, work through the tradeoffs with you and explain unfamiliar concepts. You shape the design and decide when it is ready to build; Claude then writes the code and explains what it changed and why. It installs from the author's own plugin marketplace.

2026-10-04 · SUNDAY · 13:04 PDT

  1. 1 antirez/ds4 +219 STARS TODAY · ★ 23,390 · C ds4 is DwarfStar, a self-contained local inference engine written in C for running a few strong open-weight models on hardware people actually own. It targets DeepSeek V4 Flash and V4.1 Flash first, plus GLM 5.2 and 5.3, DeepSeek V4 PRO and Qwen3.8 Flash Next, across Metal, CUDA and ROCm. It is not a general GGUF runner: the project builds and ships its own GGUF files.
  2. 2 garrytan/gstack +121 STARS TODAY · ★ 135,113 · TYPESCRIPT gstack packages the Claude Code setup used by Y Combinator president Garry Tan as 23 opinionated tools standing in for roles such as CEO, designer, engineering manager, release manager, doc engineer and QA. The MIT-licensed TypeScript repository makes its case with the author's own shipping figures over the past 60 days. It added 121 stars today.
  3. 3 dzhng/jevgrep ★ 2,248 · TYPESCRIPT jevgrep is a command-line code search tool for coding agents: ask a repository a question in plain language and jg returns relevant files, reading leads and verbatim source excerpts in one response. It uses the Jev model to judge relevance across folders, files and declarations. The project reports completing the same 8 of 10 SWE-bench tasks as its baseline at about 30 percent lower cost.
  4. 4 unreallabsai/unreal-agent ★ 2,064 · GO Unreal Agent is an async-first agent harness in Go, built around a coordinator that runs LLM turns and dispatches tool work as asynchronous operations. Sessions are append-only persisted histories that can be forked, and inputs carry caller-supplied ids that stay stable across redeliveries so duplicates are dropped. The repository ships the library, executables and benchmark runners together.
  5. 5 feitangyuan/onetake ★ 1,499 · PYTHON onetake is a Claude Agent Skill for making product launch films, teasers and feature demos as one continuous take instead of a stack of scenes separated by cuts. Each beat is built to grow out of the one before, and a verification script scores continuity at every boundary. The repository ships ten case films and uses a PolyForm Noncommercial license.

2026-10-03 · SATURDAY · 13:09 PDT

  1. 1 affaan-m/ECC +954 STARS TODAY · ★ 272,111 · JAVASCRIPT ECC packages agent-harness tuning into one installable system for Claude Code, Codex, OpenCode and Cursor. It ships skills, instincts, memory, security and research-first development workflows as a plugin plus the ecc-universal and ecc-agentshield npm packages. The repo is MIT-licensed, with a hosted GitHub App sold separately for private repos.
  2. 2 calesthio/OpenMontage +328 STARS TODAY · ★ 62,665 · PYTHON OpenMontage turns a coding assistant into a video production studio built around 12 production pipelines and more than 100 tools. It carries over 700 agent skill and production-knowledge files and wires in ElevenLabs, FFmpeg and Flux for voice, editing and image generation.
  3. 3 addyosmani/agent-skills +305 STARS TODAY · ★ 100,794 · JAVASCRIPT This skill pack encodes the workflows and quality gates senior engineers use, so that coding agents follow them consistently. Nine slash commands map to the development lifecycle, from /spec and /plan through /test, /review and /ship, and skills also fire automatically by task. A /build auto mode plans and implements every task in one approved pass, committing each individually.
  4. 4 thedotmack/claude-mem +218 STARS TODAY · ★ 95,485 · TYPESCRIPT claude-mem gives coding agents memory between sessions by capturing what the agent does, compressing the record with AI and injecting the relevant parts back as context later. It supports Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot and OpenCode, and stores the compressed history in ChromaDB for retrieval.
  5. 5 anthropics/claude-code +127 STARS TODAY · ★ 149,165 · TYPESCRIPT Anthropic's agentic coding tool lives in the terminal, reads the surrounding codebase and acts on natural-language commands. It executes routine tasks, explains unfamiliar code and handles git workflows. The repository serves as the public issue tracker and release channel for the CLI.

2026-10-02 · FRIDAY · 13:13 PDT

  1. 1 Panniantong/Agent-Reach +683 STARS TODAY · ★ 88,406 · PYTHON A Python command-line tool that gives an AI agent read and search access to Twitter, Reddit, YouTube, GitHub, Bilibili and XiaoHongShu without paying platform API fees, and that keeps picking, installing and health-checking the access route as those routes change. It added about 683 stars today, the largest daily gain in the set, taking it near 88,400.
  2. 2 coreyhaines31/marketingskills +139 STARS TODAY · ★ 52,349 · JAVASCRIPT A library of agent skills for marketing work - conversion optimization, copywriting, SEO, analytics and growth engineering - usable from Claude Code, Codex, Cursor, Windsurf or any client that supports the Agent Skills spec. The MIT-licensed collection is funded by disclosed partner integrations listed beside neutral options. About 139 stars added today.
  3. 3 microsoft/VibeVoice +36 STARS TODAY · ★ 54,601 · PYTHON Microsoft's open-source voice stack, spanning long-form text to speech and the VibeVoice speech recognition line. Its most recent release, dated September 3, is a streaming recognizer that transcribes who said what as speech arrives, with custom hotwords and 10 languages; an earlier BitNet build compresses the model to 1.58 GB for real-time CPU inference.
  4. 4 Tracer-Cloud/opensre +23 STARS TODAY · ★ 11,343 · PYTHON An open-source framework for building AI site-reliability agents, together with the training and evaluation environment they need to improve. It connects to more than 60 tools a team already runs, lets them define their own workflows, and answers production questions against their own infrastructure. Apache 2.0, and labelled public alpha at v0.1.
  5. 5 kaankiziltug/logo-design-skill ★ 1,500 · HTML An agent skill that turns Claude, Gemini CLI, Codex or Cursor into a disciplined identity designer, from the first brief to production SVG files and brand guidelines. It ships dependency-free Python tools that audit an SVG, build 16-pixel, single-color, reversed and competitor shelf test sheets, and export favicon and app-icon sets, plus a classified library of more than 1,400 real-world logos.

2026-10-01 · THURSDAY · 13:12 PDT

  1. 1 earendil-works/pi +294 STARS TODAY · ★ 111,085 · TYPESCRIPT An agent toolkit that packs a unified LLM API, an agent loop, a terminal UI and a coding-agent CLI into one TypeScript project. It picked up about 294 stars today on a base of 111,085, the largest gain in the trending pool.
  2. 2 tile-ai/tilelang +157 STARS TODAY · ★ 8,057 · PYTHON A domain-specific language for writing high-performance GPU, CPU and accelerator kernels, meant to cut the boilerplate out of hand-tuned kernel code. The Python project added about 157 stars today, taking it to 8,057.
  3. 3 PostHog/posthog +31 STARS TODAY · ★ 40,080 · PYTHON The open-source product analytics platform, now bundling AI observability, session replay, feature flags, experiments and error tracking in one developer toolkit. Its pitch is that the context behind the dashboards is also what coding agents need to diagnose a problem, steerable from Slack, web, desktop or MCP.
  4. 4 KKKKhazix/AIHOT ★ 4,606 · TYPESCRIPT A self-hosted TypeScript framework for a site that finds its own trending stories and writes its own daily digest. Swap in your own sources and selection criteria and it becomes a trending page for any industry; the stack is built on RSS feeds, LLM curation and MCP. The repository was created on September 28 and has 4,606 stars.
  5. 5 yetone/magpie ★ 4,062 · GO A macOS menu bar app that points coding agents at whichever model you pick, so Codex can run on DeepSeek or Claude Code on Kimi. The Go project covers Claude Code, Codex and Gemini CLI, was created on September 23 and has 4,062 stars.

2026-09-30 · WEDNESDAY · 13:10 PDT

  1. 1 DietrichGebert/ponytail +865 STARS TODAY · ★ 149,011 · JAVASCRIPT Ponytail makes an AI coding agent behave like a senior developer who would rather not write code at all, on the premise that the best code is the code you never wrote. It ships as agent skills and a Claude Code plugin, with rules for Cursor and other agents, and takes YAGNI as its organizing principle. The MIT-licensed repo added 865 stars today.
  2. 2 ifixai-ai/iFixAi +250 STARS TODAY · ★ 17,306 · PYTHON iFixAi audits AI agents independently, run either by a person or by the agent on itself, to check whether an agent is doing what it is supposed to do. The Apache-2.0 command-line tool returns that answer in under 120 seconds and covers prompt injection, hallucination detection and risk assessment. Its topics map the checks onto the EU AI Act, ISO 42001, the NIST AI RMF and OWASP LLM.
  3. 3 colbymchenry/codegraph +159 STARS TODAY · ★ 72,545 · C Codegraph builds a pre-indexed knowledge graph of a codebase and keeps it in sync as the code changes, so a coding agent can query structure instead of reading files. It runs entirely locally and plugs into Claude Code, Codex, Gemini, Cursor, OpenCode, Kiro and Copilot, which the author says cuts both token use and tool calls. The MIT-licensed project is written in C.
  4. 4 openclaw/openclaw +136 STARS TODAY · ★ 390,940 · TYPESCRIPT OpenClaw is a cross-platform personal AI assistant that its maintainers pitch as an agent that really does things rather than only answering. The TypeScript project stresses running on any operating system and keeping the user's data on the user's own side. It is by far the largest repository in today's set at about 391,000 stars, with 136 added today.
  5. 5 mksglu/context-mode +88 STARS TODAY · ★ 24,421 · TYPESCRIPT Context-mode narrows what reaches a coding agent's context window, sandboxing tool output for what the project claims is a 98 percent reduction, persisting session memory and enforcing routing rules. It works through MCP servers and hooks across 17 agent platforms, among them Claude Code, Codex, Cursor, Kiro and Zed.