Skip to content

auronsun/argus

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Argus

Watch every market with a hundred eyes.

An open-source AI investment committee for US stocks, A-shares, and Hong Kong stocks. Six specialised analyst agents — including a dedicated Flow Analyst that reads insider trades, short interest, northbound flow, and the dragon-tiger list — debate a ticker in real time and a Chief Investment Officer synthesises the verdict, all streamed to a glass-morphism web UI.

License: MIT Python 3.11+ React 18 FastAPI Tailwind v4

Quickstart · Why Argus · The Committee · Stock deep-dive · Architecture · 中文


Argus dashboard — six analysts, one ticker
NVDA · $1,118.20 · +0.34%

┌─ Technical Analyst ──────────────── conviction 4/5 · bullish
│  RSI(14)=72 overbought, but golden cross intact since Nov;
│  MACD histogram flattening — short-term caution.
├─ Fundamental Analyst ────────────── conviction 3/5 · neutral
│  Fwd P/E 33 vs 5y avg 28 is rich; data-centre mix expanding,
│  earnings revisions still positive. 56% margin unprecedented.
├─ Sentiment Analyst ──────────────── conviction 4/5 · bullish
│  News flow re-affirms AI capex cycle; semis ETF inflows
│  sustained; narrative crowdedness rising.
├─ Macro Strategist ───────────────── conviction 4/5 · tailwind
│  Hyperscaler 2025 capex guidance keeps NVDA central.
│  Risks: export controls, sovereign capex pullback.
├─ Risk Manager ───────────────────── risk score 3/5
│  ATR(14)=$42 → 1×ATR stop ~$1,076; sizing 1.5–2.5% of book;
│  beta 1.7 means de-rate risk in a market drawdown.
├─ Flow Analyst ──────────────────── conviction 4/5 · accumulating
│  Form 4: 3 insider buys, 0 sells in 30d (CFO + 2 directors,
│  ~$4.2M net). Short interest 1.8% of float, DTC 1.2 — low
│  squeeze risk. Smart money quietly accumulating.
└─ Chief Investment Officer ─────────────────────────────────
   ACCUMULATE  ·  conviction 4/5  ·  horizon 3-6M
   Trend intact and macro setup structurally supportive, but
   technical extension argues for staged entry.
   Entry $1,070-1,095   Stop below $1,030
   Risks: hyperscaler capex pause · export-control escalation
          · AI-narrative crowdedness

Representative output. Argus generates this live, token-by-token, in the browser.


Why Argus

Most retail stock tools answer "what's this stock worth?" with one signal — a P/E ratio, an RSI reading, an analyst rating.

Argus answers it the way a real desk does: by debate.

You give it a ticker. Six LLM agents — each playing a distinct role on a buy-side desk — read independent slices of the data, write their independent takes, and disagree on purpose. A Chief Investment Officer agent then resolves the conflict and produces a single recommendation with conviction, horizon, entry zone, and stop zone.

The whole deliberation streams to your screen, token by token. It looks the way fintech should look.

Argus Yahoo / TradingView A single ChatGPT prompt
Multi-perspective analysis 6 specialised analysts + CIO synthesis One blended view One blended view
Smart-money flow signals Insider trades · short interest (US) · Northbound flow · Dragon-Tiger list (A-share) — read by a dedicated Flow Analyst agent Not exposed Not surfaced
Markets US · A-share · HK in one workstation Per-market silos Whatever it can web-search
Live streaming reasoning Token-by-token SSE; you can read agents thinking n/a Single block of text
Pluggable data yfinance + akshare free; Finnhub / Tushare / Longbridge optional Locked-in vendor None
Pluggable LLM Anthropic · OpenAI · DeepSeek · Qwen · NVIDIA NIM · Ollama n/a Locked to one
Self-hosted, open source MIT, runs on your laptop Cloud SaaS Cloud SaaS

The Committee

Each ticker is read by a panel of LLM personas, then synthesised:

Agent Role
Technical Analyst RSI, MACD, Bollinger, KDJ, ATR, OBV — cites specific readings
Fundamental Analyst P/E · P/B · P/S · growth · margin · moat vs sector
Sentiment Analyst News headlines, narrative shifts, crowdedness
Macro Strategist Sector + macro regime (rates, FX, AI capex, China demand, geopolitics)
Risk Manager Volatility, beta, drawdown risk, position sizing, stop zone
Flow Analyst Insider transactions · short interest (US) · Northbound flow · Dragon-Tiger list (A-share) — what smart money is doing
Chief Investment Officer Resolves disagreements → action · conviction · horizon · entry · stop · key risks

The committee panel and the synthesised verdict, live in the browser:

AI Investment Committee — six analyst agents streaming reasoning live

CIO verdict card — action, conviction, horizon, entry / stop, key risks

Three markets, one workstation

Market Coverage Data source (default) Search example
US NYSE · NASDAQ yfinance AAPL, Apple
China A-shares Shanghai · Shenzhen · STAR · ChiNext akshare 600519, 茅台
Hong Kong HKEX yfinance + akshare 0700.HK, Tencent, 腾讯

The same Stock page, the same indicators, the same committee — across all three.

Bring your own LLM

The first configured key wins, in priority order. UI-stored keys override .env.

Provider Default model Notes
Anthropic claude-opus-4-7 Best agent reasoning quality
OpenAI gpt-5.5 Latest flagship
DeepSeek deepseek-v4-pro Strong, cost-effective
Qwen (DashScope) qwen-plus Good for Chinese-market commentary
NVIDIA NIM minimaxai/minimax-m2.7 Free tier on build.nvidia.com
Ollama (local) llama3.1 100% offline, BYO model

Open Settings → paste a key → it works immediately. No restart.

Stock deep-dive page

Stock deep-dive: candlestick + indicators + fundamentals + news

Each ticker page combines a candlestick + volume chart with 12 technical indicators (RSI · MACD · Bollinger · KDJ · ATR · OBV …), a live fundamentals snapshot, fresh news (with a Finnhub fallback for thinly- covered US tickers), and the AI committee panel below. The Run button kicks off the six-agent debate; you can navigate away mid-run and the state stays alive when you come back.

Quickstart

Prerequisites: Python 3.11+, Node 20+. An LLM key is optional — Argus runs in demo mode without one.

git clone https://github.com/auronsun/argus.git
cd argus
cp .env.example .env                                                    # optional: paste keys

# Backend  →  http://127.0.0.1:8765
cd backend && python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn argus.main:app --reload --host 127.0.0.1 --port 8765

# Frontend →  http://127.0.0.1:5173
cd ../frontend && npm install && npm run dev

Open the UI, type AAPL, 600519, or 0700.HK in the search bar, click Run on the AI Committee panel, and watch the desk argue.

Architecture

┌──────────────────────────────────────────────────────────────────┐
│                          React + Vite UI                         │
│   Dashboard │ Stock Deep-Dive │ Screener │ Watchlist │ Settings  │
└────────────────────────┬─────────────────────────────────────────┘
                         │  REST + SSE + WebSocket
┌────────────────────────┴─────────────────────────────────────────┐
│                      FastAPI · Python 3.11                       │
│ ┌──────────────┐  ┌────────────────┐  ┌──────────────────────┐   │
│ │  Market      │  │  Analysis      │  │  Multi-Agent         │   │
│ │  Adapters    │  │  Engine        │  │  Committee           │   │
│ │ (US/CN/HK)   │  │ (indicators,   │  │ (5 analysts + CIO,   │   │
│ │  TTL cached  │  │  fundamentals) │  │  streamed reasoning) │   │
│ └──────────────┘  └────────────────┘  └──────────────────────┘   │
│        SQLite (state) + Parquet cache (OHLCV)                    │
└──────────────────────────────────────────────────────────────────┘

Long form: docs/ARCHITECTURE.md.

Sharing & export

Designed for showing your work, not for spam.

  • Stock pageCopy summary (Markdown) · Copy link
  • Verdict cardExport Markdown (one file with action / thesis / risks / entry / stop / disclaimer)
  • WatchlistExport JSON / Import JSON — share starter watchlists with friends

Every exported artefact carries the disclaimer below. Nothing leaks API keys, paths, or local config.

Scope & limitations

Argus is a single-user research tool that runs on your laptop, not a multi-tenant SaaS. A few things worth knowing up-front, in honest terms:

  • No authentication. The local HTTP API trusts whoever can reach the port. Don't bind it to a public IP. See SECURITY.md.
  • Free data has limits. yfinance and akshare are best-effort public endpoints — they rate-limit, occasionally rename columns, and can return empty for thinly-covered tickers. Argus mitigates with TTL caching and a Finnhub fallback for news; for serious use, plug in a paid feed (Tushare Pro, Finnhub real-time, Longbridge).
  • "Real-time" means 5-second polling. Argus polls the underlying adapters; it is not wired to vendor-native streaming. Quotes can lag a few seconds.
  • Free-tier LLM streams sometimes drop. NVIDIA NIM free tier in particular closes long streams mid-flight; the CIO call retries once on transport failures, and analyst cards display an explicit notice if a model returned no content.
  • What's wired today in the Flow Analyst: insider transactions + short interest for US tickers; Northbound (Stock Connect) flow + Dragon-Tiger list for A-shares. HK flow signals and 融资余额 are placeholders pending future work — when you analyse those tickers the Flow Analyst will say so explicitly rather than fabricate numbers.

If you want to run Argus somewhere other than localhost, treat that as a project rather than a config flip — the SECURITY.md hardening checklist is a starting point.

Acknowledgements

Argus stands on the shoulders of yfinance, akshare, FastAPI, Recharts, and TradingView Lightweight Charts.

License

MIT.

Argus is research and education software. Nothing it produces is investment advice. Past performance does not predict future results. Always do your own due diligence.


中文

用一百只眼睛盯每一个市场。

Argus 是一个开源 AI 投资委员会,覆盖 美股 · A 股 · 港股。 六位分工不同的 LLM 分析师(含业内独有的资金流分析师:读懂内幕交易、空头持仓(美股)、北向资金、龙虎榜(A 股))+ 一位首席投资官(CIO)实时辩论一只股票,并给出最终判断—— 全程 token 级流式呈现,UI 是质感不输付费产品的玻璃拟态界面。

为什么是 Argus

零售股票工具回答"这只股票怎么样"通常只给一个信号——一个 P/E、一个 RSI、一个评级。

Argus 用真实交易台的方式回答:让分析师们辩论

你输入一只代码,六位 LLM 分析师各自读取不同的数据切片,独立得出观点,故意会有分歧。CIO Agent 综合所有观点,给出带置信度、持有期、入场区间、止损区间的最终建议。

整个推理过程逐字流式显示。这才是 fintech 应该有的样子。

一分钟启动

git clone https://github.com/auronsun/argus.git && cd argus
cp .env.example .env
cd backend && pip install -r requirements.txt && uvicorn argus.main:app --reload --host 127.0.0.1 --port 8765 &
cd ../frontend && npm install && npm run dev

打开 http://127.0.0.1:5173,搜 茅台 / 腾讯 / NVDA——委员会立刻开始辩论。

委员会成员

角色 关注
技术分析师 RSI · MACD · BOLL · KDJ · ATR · OBV——引用具体读数
基本面分析师 PE · PB · PS · 增长 · 利润率 · 护城河 vs 行业
情绪分析师 新闻头条 · 叙事变化 · 拥挤度
宏观策略师 行业 + 宏观周期(利率 / 汇率 / AI 资本开支 / 中国需求 / 地缘)
风险经理 波动率 · Beta · 回撤风险 · 仓位区间 · 止损
资金流分析师 内幕交易 · 空头持仓(美股)· 北向资金 · 龙虎榜(A 股)——读懂"聪明钱"在做什么
首席投资官(CIO) 综合所有观点 → 操作 · 置信度 · 持有期 · 入场区间 · 止损 · 主要风险

自由切换 LLM 与数据源

  • LLM:Anthropic · OpenAI · DeepSeek · 通义千问 · NVIDIA NIM · Ollama(本地)
  • 行情:免费源 yfinance / akshare 开箱即用;Finnhub · Alpha Vantage · Tushare Pro · Longbridge 填 Key 即可启用
  • 在「设置」页粘贴 Key,无需重启——立即生效

免责声明

本软件仅供研究与教育用途。任何输出不构成投资建议,过往业绩不预示未来表现,请独立做尽职调查。

About

Open-source AI investment-committee for US stocks, A-shares, and HK stocks. Six specialised analyst agents debate a ticker in real time and a CIO synthesises the verdict — streamed to a polished web UI.

Topics

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors