Vibe-Trading
🇹🇭 ภาษาไทย
ภาพรวม
Vibe-Trading (by HKUDS — Hong Kong University of Data Science) คือ AI finance workspace ที่แปลง natural language → trading strategy ที่ executable ได้ รองรับ OpenClaw และ Claude Desktop ผ่าน MCP (~4,500 ⭐)
ความสามารถหลัก
Skills & Agent Teams
| ประเภท | จำนวน |
|---|---|
| Specialist Finance Skills (YAML) | 74 skills ใน 8 หมวด |
| Pre-Built Agent Teams | 29 teams (investment committee, equity desk, crypto, macro, quant) |
| Backtest Engines | 7 engines (A-shares, HK/US, crypto, futures, forex, cross-market) |
Output Export
- TradingView (Pine Script v6)
- TDX (通达信 / 同花顺)
- MetaTrader 5
สถาปัตยกรรม
Natural Language Input
→ ReAct Agent (5-layer context compression)
→ Tool-call batching
→ DAG-based Swarm Orchestration
→ Skill auto-discovery (74 YAML definitions)
→ Persistent cross-session memory (FTS5 full-text)
→ Output: strategy code + backtest results
MCP Integration: 22 tools → Claude Desktop, OpenClaw, Cursor
Data Sources (Zero Config)
6 data sources ไม่ต้องตั้งค่า พร้อม automatic fallback
Tech Stack
| Layer | Technology |
|---|---|
| Backend | Python 3.11+, FastAPI |
| Frontend | React 19, TypeScript, Vite |
| Agent | ReAct + DAG swarm |
| Memory | File-based + FTS5 |
| Package | pip install vibe-trading-ai |
LLM Support: 13+ providers — OpenAI, DeepSeek, Claude, Gemini, Groq, Ollama (ฟรี ไม่ต้อง API key)
Install Options
pip install vibe-trading-ai→ CLI- Docker (single command)
- MCP plugin → ใช้กับ Claude Desktop / OpenClaw ได้เลย
- ClawHub (managed skill distribution)
ความเชื่อมโยง
- OpenClaw — รองรับผ่าน MCP + ClawHub
- Multi-Agent Trading — concept หลัก
- TradingAgents — alternative multi-agent framework
- Algorithmic Trading — broader context
- TradingView MCP — alternative TradingView integration
🇬🇧 English
Overview
Vibe-Trading by HKUDS is a natural-language-to-strategy AI trading workspace. Users describe trading ideas conversationally; the agent generates, backtests, and exports executable code (~4,500 ⭐). Integrates natively with OpenClaw and Claude Desktop via MCP.
Key Numbers
- 74 specialist finance skills across 8 categories
- 29 pre-built agent teams (investment committee, equity desk, crypto, macro, quant)
- 7 backtest engines covering A-shares, HK/US equities, crypto, futures, forex, cross-market
- 22 MCP tools exposed for external AI clients
- 13+ LLM providers supported (including Ollama with no API key)
Architecture
ReAct agent with 5-layer context compression + DAG-based swarm orchestration. Skills are auto-discovered from 74 YAML definitions. Memory persists across sessions via FTS5 full-text indexed file storage.
Export Targets
TradingView Pine Script v6, TDX (通达信/同花顺), MetaTrader 5.
Related Pages
- OpenClaw — MCP + ClawHub integration
- Multi-Agent Trading — Concept
- TradingAgents — Alternative framework
- TradingView MCP — TradingView-specific MCP