Back home

MineAgent

github 0.52
Visit source

Intro

MineAgent is a local open project for visual Agent orchestration: design industrial-style Agent flows on a browser canvas (React Flow), pick a framework to auto-generate an architecture diagram, edit nodes/code/models, then Save to compile the canvas into a standalone runnable Python Agent project, and Run to execute on your machine with streaming per-node feedback. Repo: https://github.com/YT-Proteus-bot/MineAgent (created 2026-11-05, ~111★). README tagline (ZH): agent customization + visual flows to lower the barrier for everyday users; commercial use should credit the source.

Stack (from README):

  • Frontend: React 18 + TypeScript + React Flow + Vite;
  • Backend: Python FastAPI + SQLite (no ORM / no extra DB deps claimed);
  • Storage: canvases in SQLite; generated code under projects/<canvas>/;
  • Start: start.bat / start.ps1 → default http://127.0.0.1:8000.

Explain

Problem

Writing multi-agent pipelines by hand (LangChain/LangGraph, AutoGen, CrewAI, …) is steep for non-specialists. MineAgent aims to make flow design visual, then emit real Python you can run and extend.

Capabilities (README)

  • Canvas: drag/connect ~110 node types (agents, tools, MCP, retrieval, guards, human approval, …); constraints panel with built-in rules;
  • Reasoning modes per agent (e.g. ReAct, Plan-and-Execute, Goal, Reflexion, ToolCall, Simple) plus org relations (superior/subordinate/peer) with ctx.delegate;
  • Streaming run: active node highlights; token-level model output in the run panel;
  • Community templates via GitHub PR upload/load; built-in AI assistant panel for canvas Q&A;
  • Optional tools/mock_llm.py so the full canvas→code→run loop works without a real API key.

Limits

  • Oriented to local Windows-friendly startup scripts; verify platform support for your OS;
  • API keys are stored encrypted locally for node/model config—still a secrets-handling responsibility;
  • Star count and feature surface will change; treat README as source of truth.

Usage

Who it's for

  • Builders who want a visual canvas that compiles to real Python Agent projects;
  • Teams prototyping multi-agent / RAG / tool flows across common frameworks;
  • Educators demonstrating Agent orchestration without starting from a blank repo.

Getting started

  1. Clone https://github.com/YT-Proteus-bot/MineAgent and follow the README quick start (start.bat or .\start.ps1);
  2. Open http://127.0.0.1:8000, pick a framework / reasoning-mode template, edit the canvas, Save to generate projects/<name>/;
  3. Configure models under user center (or run tools/mock_llm.py for a dry end-to-end path);
  4. Click Run to stream node execution; use the constraints panel before productionizing generated code.

Advertisement