Other people's great work, filtered by Laojin
GitHub holds thousands of AI projects — what beginners lack isn't lists, it's filtering. Only repos and tools Laojin personally verified, proven by time or by numbers, make this board. Small, sharp, continuously updated. Our own 365 products never appear here.
Stars are a 2026-07 snapshot for scale reference only; see GitHub for live numbers.
MCP Servers（官方合集）
⭐ 88.7k · 2026-07 snapshot
The official entry point for connecting agents like Claude or Cursor to real data and tools — filesystem, search, databases, Slack, all with reference implementations. If you're learning what a good MCP server looks like, copy the official ones.
Why it's pickedAwesome Claude Code
⭐ 50.5k · 2026-07 snapshot
A one-stop, community-maintained list for Claude Code users: workflows, hooks, slash commands, CLAUDE.md templates and MCP configs. Everything you need to take Claude Code from 'works' to 'fits like a glove'.
Why it's pickedn8n
⭐ 197.2k · 2026-07 snapshot
A workflow-automation platform non-programmers can actually use, with native AI nodes. For the repetitive flows on your desk — syncing data, scheduled notifications, form handling — this should be your first stop.
Why it's pickedAutoGPT
⭐ 185.6k · 2026-07 snapshot
The project that kicked off autonomous agents. Its value today is the platform shape: an agent marketplace, reusable blocks, triggers. The most complete specimen if you want to see where fully-autonomous agents hit product walls.
Why it's pickedCrewAI
⭐ 55.9k · 2026-07 snapshot
The fastest on-ramp among multi-agent frameworks: write each agent a job description (role, goal, backstory) and they work as a crew. Even founders from a business background can follow its design logic.
Why it's pickedAutoGen
⭐ 59.9k · 2026-07 snapshot
Microsoft's multi-agent conversation framework, credible in both research and engineering: built for complex tasks where agents must discuss, review and debate to produce a result — code review, proposal critique and the like.
Why it's pickedDify
⭐ 149.5k · 2026-07 snapshot
For companies that want AI apps fast without writing everything: visual orchestration, RAG, dozens of model providers, access control — all self-hostable. The default base for an SME's AI proof-of-concept.
Why it's pickedFirecrawl
⭐ 153.6k · 2026-07 snapshot
Feed any website to an LLM cleanly: crawl, denoise, convert to markdown in one API call. The most frictionless 'web → corpus' plumbing for content pipelines, competitor monitoring and knowledge-base ingestion.
Why it's pickedAwesome LLM Apps
⭐ 125.2k · 2026-07 snapshot
100+ runnable AI agent and RAG applications, all with code and docs. Clone one close to your business and adapt it — ten times faster than a blank page, and the best material for learning how real apps are assembled.
Why it's pickedOllama
⭐ 176.5k · 2026-07 snapshot
The lowest bar for running open models locally: one command brings up Llama, Qwen or DeepSeek, and private data never leaves your machine. Always the first step in testing 'is a local model good enough for this'.
Why it's pickedllama.cpp
⭐ 121.1k · 2026-07 snapshot
The engine under local inference: Ollama and countless desktop AI apps are built on it. Unavoidable when you need models on low-spec machines or edge devices, or fine control over quantization and VRAM.
Why it's pickedGenerative AI for Beginners
⭐ 113.3k · 2026-07 snapshot
Microsoft's 21-lesson generative AI course: progressive from concepts to hands-on, with code and videos. The most standard starting line worldwide for learning how AI applications are built.
Why it's pickedPrompt Engineering Guide
⭐ 76.8k · 2026-07 snapshot
An encyclopedic guide to prompting: from basics through CoT, RAG and agent-related papers, with Chinese translations available. The one course to take if you want noticeably better output from AI.
Why it's pickedLLM Cookbook（吴恩达课程中文版）
⭐ 24.4k · 2026-07 snapshot
The Chinese edition of Andrew Ng's LLM course series, maintained by the Datawhale community: eleven courses covering prompting, RAG, evaluation and agents. The smoothest textbook for Chinese-speaking developers entering LLM application development, bar none.
Why it's pickedAnthropic Agent Skills（官方）
⭐ 163.0k · 2026-07 snapshot
Anthropic's official Agent Skills repo — the de facto reference set for the SKILL.md format: document processing, design and office automation with official implementations. Agent skills are moving from a Claude-only feature to an industry-wide convention.
Why it's pickedOpenAI Agents SDK
⭐ 28.0k · 2026-07 snapshot
OpenAI's official Agents SDK: lightweight, built-in tracing, and a clean handoff mechanism. The shortest path to production agents if you're already in the OpenAI ecosystem.
Why it's pickedGitHub Spec Kit
⭐ 122.9k · 2026-07 snapshot
GitHub's official Spec-Driven Development toolkit: write the spec (spec/plan/tasks) first, then let AI generate the code — pulling vibe coding back into engineering discipline. The key remedial step as AI coding moves from toy to production.
Why it's pickedGemini CLI
⭐ 106.1k · 2026-07 snapshot
Google's open-source terminal agent: a generous free tier, built-in Google Search and MCP support. A capable assistant right in your shell — and the best second knife to keep beside Claude Code.
Why it's pickedChrome DevTools MCP
⭐ 47.3k · 2026-07 snapshot
Google's official MCP that lets coding agents drive Chrome directly — inspect the DOM, read console errors, run performance traces. The missing piece for having AI verify frontend work in a real browser itself.
Why it's pickedCoze Studio（扣子开源版）
⭐ 21.2k · 2026-07 snapshot
ByteDance's open-sourced agent development platform: visual orchestration, a plugin ecosystem and knowledge bases, with a Chinese-friendly experience. A ready-made foundation if your company wants its own 'Coze-like' internal AI platform.
Why it's pickedSuperpowers（Agent 技能框架）
⭐ 258.4k · 2026-07 snapshot
The most influential agent-skills collection in the community: obra turned a full software-development methodology — TDD, plan-first, systematic debugging, code review — into installable SKILL.md skills. Once installed, Claude Code's discipline visibly improves.
Why it's pickedskill-creator（Anthropic 官方造技能技能）
⭐ 163.0k · 2026-07 snapshot
The official 'skill for creating skills' inside Anthropic's repo: tell the agent about a process you want to capture and it walks you through producing a standard SKILL.md — frontmatter, trigger description, steps and resource files. The first brick for any team starting its own skill library.
Why it's pickedwshobson/agents（子代理技能市场）
⭐ 38.1k · 2026-07 snapshot
The largest community subagent collection: dozens of role-based Claude Code / Codex plugins — security audit, performance analysis, architecture review, per-language specialists — installable out of the box, covering most professional coding-agent scenarios.
Why it's pickedClaude Code Templates（aitmpl）
⭐ 29.8k · 2026-07 snapshot
A one-stop template library plus CLI for Claude Code: agents, commands, hooks and MCP configs organized by framework and language, installed into your project with one command. The fastest cold-start for Claude Code on a new project.
Why it's picked