🎯 一句話 verdict: Crawl4AI 係 2026 年 GitHub #1 trending repo · 73.2k stars · #1 開源 LLM-friendly web crawler。直接解決你做 research 嗰陣嘅痛點 — 你而家用 web_extract 嘅嘢,Crawl4AI 自己 host + 100% local + 自帶 LLM extraction。對你 Hermes stack 嘅即時 value 係:你再唔使靠 web tool quota。
當前版本:v0.9.2(2026/7/15 release · 4 日前) · License:Apache-2.0 · 完全免費開源 · Python 98.8%。
📊 一、5 大關鍵數字
| 指標 | 數值 |
|---|
| GitHub Stars | 73,200 |
| Forks | 7,500 |
| Used by | 3,200+ repos |
| Contributors | 87 |
| Latest release | v0.9.2(2026/7/15,4 日前) |
| Commits | 1,589 |
| Branches / Tags | 178 / 59 |
| Language | Python 98.8% |
| GitHub #1 trending | ✅ |
5 大目標
- Generate Clean Markdown — 完美 for RAG pipelines / 直接 ingest LLM
- Structured Extraction — CSS / XPath / LLM-based extraction
- Advanced Browser Control — hooks / proxies / stealth modes / session re-use
- High Performance — parallel crawling / chunk-based / real-time
- Open Source — 冇 forced API key / 冇 paywall
一句話定位
將任意 web 頁面轉成clean structured markdown + LLM-friendly format,100% open source、完全 local run。
🛠️ 三、技術架構(推測 / 從 code 觀察)
Stack
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
| ┌──────────────────────────────────────────┐ │ AsyncWebCrawler (Python async) │ ├──────────────────────────────────────────┤ │ Browser Engine (Playwright) │ │ • Multi-browser support │ │ • Stealth mode (anti-bot) │ │ • Session re-use │ ├──────────────────────────────────────────┤ │ Content Processing │ │ • Markdown generation │ │ • Citation extraction │ │ • BM25 / pruning filters │ │ • Links / images / tables / metadata │ ├──────────────────────────────────────────┤ │ Extraction Strategies │ │ • JsonCss / JsonXPath / JsonLxml │ │ • Regex / Cosine similarity │ │ • LLM-based extraction │ ├──────────────────────────────────────────┤ │ Deep Crawl │ │ • BFS / DFS / BestFirst │ │ • URL filters + scorers │ ├──────────────────────────────────────────┤ │ Browser Management │ │ • Lifecycle / viewport / wait_for │ │ • Stealth / sessions / iframes │ └──────────────────────────────────────────┘
|
Repository 結構
1 2 3 4 5 6 7 8 9
| crawl4ai/ # Main library deploy/docker/ # Docker server docs/ # Documentation prompts/ # LLM prompts library sbom/ # CycloneDX SBOM scripts/ # Stats dashboard etc tests/ # 291+ regression tests .claude/ # Claude AI integration .context/ # HTTP strategy / file download
|
💻 四、最簡 Quick Start(5 行)
1 2 3 4 5 6 7 8 9
| import asyncio from crawl4ai import AsyncWebCrawler
async def main(): async with AsyncWebCrawler() as crawler: result = await crawler.arun(url="https://crawl4ai.com") print(result.markdown)
asyncio.run(main())
|
安裝
1 2 3 4 5 6
| pip install crawl4ai==0.9.2
docker pull unclecode/crawl4ai:0.9.2 docker pull unclecode/crawl4ai:latest
|
✨ 五、6 大 Killer Features
1. Adaptive Web Crawling
用 advanced information foraging algorithms,自動知道幾時停(已經搵到足夠 info answer 問題)。
- No-LLM: JsonCss / JsonXPath / JsonLxml / Regex(平)
- LLM-based: 用任何 LLM extract(quality 高)
- Cosine similarity: 內容相似度過濾
- Clustering: Auto group similar pages
3. Browser Stealth + Session Re-use
- Anti-bot detection
- Persistent sessions(cookie / login state 保留)
- Proxy rotation
4. Parallel + Chunked
- Multi-page concurrent crawling
- Chunk-based extraction(適合 long content)
- Real-time streaming
5. RAG-Optimized Output
- Clean markdown(noise-free)
- Citation tracking
- BM25 / pruning filter
- Metadata extract
6. Docker Server + Adaptive Crawling Skill
- Self-hosted Docker server
- Pre-built AI skill for Claude / Cursor / Windsurf(23K+ words SDK ref)
📦 六、v0.9.2 最新變動(2026/7/15 release)
| 改動 | 細節 |
|---|
| Dispatcher 修復 | Clean up leaked tasks/pages on streaming crawl close (#2071) |
| Docker Playground 修復 | Advanced Config 400 error (#2059) |
| Docker Monitor WebSocket 修復 | 500 error (#2060) |
| Playwright headless shell | Bundled in Docker image (#2067) |
| GPU build 修復 | ENABLE_GPU=true CUDA toolkit support (#2020) |
| Breaking changes from 0.9.0 | 詳細 migration guide in deploy/docker/MIGRATION.md |
0.9.0 重大 security redesign(2026/6)
- R1–R7 secure-by-default Docker hardening
- 3 大 security fixes: download path traversal、streaming-path SSRF、extra_args RCE
- Uniform egress pinning 所有 browser-fetch endpoints
- 309 security tests passed
🆚 七、Competitor 對標
| Tool | 對標 | 強項 | 弱項 |
|---|
| Crawl4AI | Open-source LLM-friendly | ✅ 完全免費 · ✅ self-hosted · ✅ 73.2k stars | ⚠️ 需要 Python + browser setup |
| Firecrawl | Cloud LLM scraping | ✅ 簡單 · ✅ managed | ❌ 收費(~$0.50/1k pages) |
| Playwright | Browser automation | ✅ 全功能 · ✅ 通用 | ❌ 唔專注 LLM extraction |
| Scrapy | Web crawler framework | ✅ 老牌 · ✅ scalable | ❌ 唔 LLM-friendly |
| Beautiful Soup | HTML parser | ✅ 簡單 · ✅ 普及 | ❌ 唔 render JS |
| Jina Reader | URL → markdown API | ✅ 簡單 · ✅ free tier | ❌ Cloud dependency |
Crawl4AI 嘅 positioning: 唯一一個完全 self-hosted、open-source、LLM-first 嘅 production-grade crawler。
🛠️ 八、對標你 Hermes Stack — 即時機會
| # | Use Case | 你點用 |
|---|
| 1 | 取代 web_extract | 你而家每次研究都靠 web_extract 抓 Lark doc 內容。Crawl4AI self-host 後無限 quota |
| 2 | RAG pipeline 俾客戶 | 用 Crawl4AI + FAISS + LLM build RAG 賣俾 HK SME |
| 3 | Competitive intelligence | 自動 monitor 客戶 competitor 網站 pricing / news |
| 4 | Lead generation | 自動 scrape LinkedIn / industry directory → CSV |
| 5 | Content monitoring | 政府公告 / 招聘 / 法規 update monitor |
對標你 stack
| 你有 | Crawl4AI 加成 |
|---|
web_search / web_extract | ⚠️ 取代 → 你自己 host crawler,無限 quota + 更快 + 更 deep |
Lark doc 自動研究 | ✅ 完全 unblock quota limits |
n8n-automation | ✅ Build long-running scrape workflow(每週 crawl 100 頁) |
| Hermes cron jobs | ✅ Crawl4AI 你自己 host,可以 build 任何 monitor 場景 |
| MiniMax Music 3.0 | ➖ N/A |
💰 九、HK Side Income 對標
5 大 monetization 路徑
| 路徑 | 客戶 | 報價 |
|---|
| 1. RAG consulting | HK SME 想 build 內部 knowledge base | HK$30k–100k |
| 2. Competitive intelligence dashboard | Marketing team | HK$15k–40k/月 |
| 3. Lead generation pipeline | B2B sales team | HK$10k–30k/單 |
| 4. Government data scraping | Policy researcher / NGO | HK$20k–60k |
| 5. AI skill bundle | 賣 Crawl4AI 教學 + n8n + RAG template | HK$5k–15k/套 |
即時 3 個 0 成本 demo
| Demo | 步驟 |
|---|
| A. RAG demo 俾客戶 | Crawl4AI → markdown → FAISS → MiniMax API |
| B. Competitor monitor | Crawl4AI + n8n cron + Lark Sheet update |
| C. SME 月報增強 | 你個 monthly cron 加 Crawl4AI 監察政策原文 |
⚠️ 十、限制 / 風險
- 需要 Python + Playwright browser — 首次 setup 要 install Chromium / Firefox
- Self-host 需要 server — Docker 跑長期要 dedicated VM / NAS
- Anti-bot detection 對最強網站 — Cloudflare / DataDome 等仍然會 block
- Rate limit 由你自己控制 — 太 aggressive 會被 IP ban
- Python 98.8% — Node.js / Go 用家要 wrap
- 0.9.0 → 0.9.2 breaking changes — 升級要睇
deploy/docker/MIGRATION.md - 73k stars but 87 contributors — Project 健康但 maintainer 集中(bus factor)
- 法律風險 — Scraping 受 ToS / GDPR 限制,商業用要小心
🚀 十一、Next Step 建議
| 你想 | 我幫你做 |
|---|
| 即刻裝 Crawl4AI | 出 macOS 安裝 SOP + Playwright setup |
| Build RAG demo | Crawl4AI + FAISS + MiniMax API → HK SME 內部 KB 範本 |
| 寫 competitor monitor | n8n + Crawl4AI + Lark Sheet 自動 update |
| 取代 web_extract | Build Hermes wrapper 用 Crawl4AI 做底層 |
| 拍 YouTube tutorial | 「點用 Crawl4AI 5 分鐘 build RAG」教學 |
📚 十二、Source / 延伸閱讀
GitHub & Code
同類工具
Comparison 文章
數據截至 2026/7/19(v0.9.2 release 4 日前)。所有技術細節以 GitHub README + 官方 docs 為準。