MCP e chamadas de ferramentas

Vibe-Trading: Your Personal Trading Agent

hkuds/vibe-trading

An open-source Python-based AI agent framework designed to provide comprehensive trading, backtesting, and quantitative finance capabilities via a single command.

★ 30,1KEstrelas
⑂ 4,8KForks
41Problemas em aberto
PythonIdioma
MITLicença
Q@project.QualityScorePontuação editorial

Capturas de tela do projeto

Captura de tela de Vibe-Trading: Your Personal Trading Agent Captura de tela de Vibe-Trading: Your Personal Trading Agent

Visão geral

Vibe-Trading is a multi-agent AI trading system developed by HKUDS. It integrates an LLM-powered agent with a FastAPI backend and a React 19 frontend, allowing users to deploy a personal trading agent for market research, backtesting, and execution. The project supports multiple global markets, including US, Canada, China (A-shares), Korea (KRX), and cryptocurrency (USD-M perpetuals). It features a robust quantlib layer with tested financial math functions, an auditable execution ledger, and an MCP (Model Context Protocol) surface exposing 64 tools.

Principais recursos

  • Multi-agent AI core with LLM integration
  • Backtesting engines for 9 markets (US, Canada, China, Korea, Crypto, etc.)
  • MCP surface with 64 tools
  • Tested `quantlib` finance-math layer (~250 functions)
  • Shadow Account for simulated tracking
  • Risk x-ray artifacts for portfolio backtests
  • Hash-chained audit ledger for tamper detection
  • Sandboxed strategy code generation (blocks broker imports, sockets, subprocesses)
  • Electron desktop shell with safeStorage credential management
  • Five-locale web UI support

Requisitos, instalação e início rápido

Install via PyPI using: pip install vibe-trading-ai

Uso

After installation, users can run the agent via command line or interact through the web UI. The system supports non-interactive runs (`vibe-trading run`), scheduled research jobs with IANA timezone support, and MCP integration. State and sessions are stored under `~/.vibe-trading` (relocatable via `VIBE_TRADING_HOME`).

Compatibilidade do modelo e casos de uso

Supports OpenAI-compatible endpoints, including ModelScope (defaulting to Qwen/Qwen3.5-27B). Includes adapters for Claude models (opus-4-7, opus-5, sonnet-5) that handle deprecated `temperature` fields. Supports OpenAI Codex OAuth and direct-OpenAI `reasoning_effort`.

Observações sobre licença e riscos

MIT License.

Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: MIT. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Lançamento e manutenção

Recent updates (as of August 2026) include secure Windows packaging, Canadian equities market support, ModelScope provider integration, a tested finance-math layer (`src/quantlib`), institutional holdings tools, Korea (KRX) market engine, and various correctness fixes for fundamentals, A-share prices, and gap-safe returns.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33,9KPython

Composio

composiohq/composio

★ 29,5KTypeScript