跳到正文
GitHub·· 9 天前AI 评分60

langplay 发布可视化工具,无需写代码搭建 LangChain 与 LangGraph 工作流

DnD-World/langplay

AI 导读

langplay 是一个可视化学习与搭建 LangChain、LangGraph 工作流的开源项目,用户以积木方式堆叠 prompt、AI、工具、文档、router、critic 等步骤,或画成带分支和循环的图,无需写代码。

正文

CI License

Learn LangChain and LangGraph without writing code: build an AI workflow from blocks, run it on a free AI, see every step, then ship it as your own tool.

Langplay: build a recipe, run it on a real AI, inspect every step

The problem

LangChain and LangGraph tutorials start with Python, API keys and a notebook. Beginners never get to see what a chain, a tool call, a router or a loop actually does — and when they finally build something, it stays stuck in a notebook.

What it does

  • Build visually. Stack steps (prompt, AI, tool, documents, router, critic, final answer) as a list or draw them as a graph with branches and loops.
  • Run on real AI for free. Free services by default, a 2-minute guide to free keys (Gemini, Groq, OpenRouter), or any OpenAI-compatible service. An offline practice mode if nothing answers.
  • See everything. Click any step to read the exact prompt sent, the answer, the router's choice, the critic's verdict, sources, time, tokens and cost.
  • Real tools, no keys. Wikipedia, web answers, an exact calculator, the clock — and search your own PDFs, read inside your browser.
  • Learn by doing. 12 auto-checked lessons, from your first chain to a critic loop, with XP and ranks.
  • Compare. Same question through two models or two recipes, side by side.
  • Spin it out. Turn a recipe into a tool page, a website widget, one HTML file, LangGraph Python, a Claude skill, an n8n workflow, or an MCP tool Claude can call.

Quick start

bun install
bun run dev:local      # http://localhost:20136

Live: https://langplay.stravelakis.com · In-app docs: /docs · Full setup: INSTALL.md · Full manual: GUIDE.md

Screenshots

Graph view with a router's two paths Spin out dialog with seven formats
A spun-out tool page answering with a real calculator Phone layout, Live Run first

Status

  • Works: everything above; 58 automated tests; spin-outs verified end to end (LangGraph 1.2, n8n, MCP over stdio, Cloudflare Worker locally).
  • Limits: free keyless AI services allow only a few requests (Pollinations asks for payment after a few; OVHcloud ~2 a minute) — a free key is the reliable path. Progress, recipes and documents live in the browser only; no accounts.
  • Next: see HANDOFF.md.

Contributing

Issues and PRs welcome. Please read the Code of Conduct. Security problems: see SECURITY.md.

License

MIT. Third-party credits in NOTICE.

来源:GitHub · github.com