Most retail traders and small research teams lack an automated pipeline that reliably merges live market quotes, structured fundamentals, news/sentiment, and multi-step LLM reasoning into a concise, actionable daily report. This repository fills that gap by turning heterogeneous market feeds and search results into a reproducible, LLM-driven "decision dashboard" that can run on a schedule and deliver multi-channel notifications.
What Sets It Apart
- LLM-first decision dashboard: instead of only outputting raw indicators, the system prompts an LLM to synthesize technical signals, chip distribution, fundamentals and news into a short actionable conclusion with buy/hold/sell guidance and explicit checklists. So what: it reduces the manual effort of reading multiple sources and produces a compact operational summary for daily routines.
- Multi-source, fail-open pipeline: integrates AkShare/Tushare/Pytdx/YFinance for quotes, several web search APIs (SerpAPI, Tavily, Brave/MiniMax) for news, and optional social sentiment APIs for US stocks. So what: missing feeds degrade gracefully (fail-open) to keep the analysis running rather than failing entirely.
- Agent & strategy skills: built-in Agent modes and 11 strategy skills (moving average, Chanlun, wave theory, Regime Strategy, etc.) let users run multi-step strategy dialogues and backtest-like validations. So what: you can both automate daily signals and interactively ask the system for strategy reasoning without leaving the platform.
- Zero-cost scheduling options: designed to run in GitHub Actions or Docker, with extensive environment-variable driven configuration for LLM channels, notification targets and report templates. So what: small teams can run scheduled analyses without hosting costs.
Key capabilities (compact)
- Generates per-stock and market-level dashboards with a headline conclusion, exact entry/stop/target points, and an operational checklist.
- Supports A-share, HK and US markets plus US indices; supports vision-based import (image → stock list) and CSV/clipboard import.
- Multi-channel notifications: corporate WeChat, Feishu, Telegram, Discord, Slack, email and more, including options to render Markdown to image for channels that need it.
- Model-agnostic LLM integration: configurable to use Gemini/Anthropic/OpenAI/AIHubMix/Ollama/local models via LiteLLM or OpenAI-compatible endpoints; includes multi-key/fallback and rate-conservative settings.
Who it's for — tradeoffs and limitations
Great fit if you are an individual trader or small team who wants a reproducible daily analysis pipeline that blends data, news and LLM reasoning and can be run with minimal infra cost (GitHub Actions / Docker). It’s also useful for developers wanting a modular base to extend strategies or add notification channels. Look elsewhere if you need a fully audited, latency-sensitive execution platform (this project is for analysis and notifications, not automated order execution), or if you require institutional-grade market data SLAs—third-party data freshness and LLM outputs depend on configured providers and entail API costs. Also, outputs are intended as research/decision support and not financial advice.
Where it fits
This project sits between simple signal scripts and full commercial analytics platforms: it’s heavier than single-indicator bots because it fuses news + LLM synthesis, but lighter and more deployable than enterprise-grade terminals because it’s open-source, GitHub-Action friendly, and designed to be extensible by users who accept the external-data and LLM reliability tradeoffs.
