simonlin1212/investment-news

Investment News – A‑share industry‑signal dashboard

What it is – A locally‑run Python tool that fetches ≈ 100 public news/RSS sources, runs a large‑language model (your own Claude subscription or any OpenAI‑compatible API) to extract 3‑5 Chinese “today’s‑key‑points” per sector, translates the original English headlines, and displays everything on a single browser page (http://localhost:8793).

Why it matters – It combines three pieces of technology that are often separate:

1. Web‑scraping of many international media outlets (pure standard‑library code, no third‑party packages).
2. LLM‑powered summarisation & translation performed locally, so no data leaves your machine.
3. A sector‑focused UI that maps global industry news to the 12 A‑share sectors most relevant to Chinese investors.

---

Core capabilities (as listed in the README)

Feature	What you get
今日要点 (Daily key points)	3‑5 concise Chinese bullet points per sector, deduplicated across sources, generated by the LLM.
双语呈现 (Bilingual view)	English headline shown alongside its Chinese translation; original article link is attached.
12 sector coverage	AI/large‑model, semiconductor, robotics, EV, energy, biotech, aerospace, cybersecurity, internet/tech, consumer electronics, macro‑economics, science.
一键刷新	Click the refresh icon in the dashboard; the server runs a fetch‑then‑digest pipeline and reloads the page automatically.
双模型选项	Use a local claude-cli subscription (free after login) or any OpenAI‑compatible API (DeepSeek, OpenAI, OpenRouter, etc.).
本地运行 & 零依赖	All fetching, processing, and serving are done with Python’s standard library; no database, no external hosting.
合规过滤	Built‑in keyword blacklist removes gambling, crypto‑trading, porn, etc.; political and financial news are kept.

---

Quick start (from the README)

# clone & enter repo
git clone https://github.com/simonlin1212/investment-news.git
cd investment-news

# configure LLM (optional – defaults to local Claude subscription)
# edit llm.config.json if you prefer an OpenAI‑compatible API

# start the local server (default port 8793)
python3 server.py

# open the dashboard
# macOS: open http://localhost:8793
# Windows: start http://localhost:8793
# Linux: xdg-open http://localhost:8793

Click the circular refresh button in the top‑left corner to trigger a new fetch‑and‑summarise run.

---

How it works (pipeline diagram from the README)

1. sources.json – Lists 108 RSS/JSON endpoints, each tagged with a sector (hint).
2. scripts/fetch.py – Downloads the feeds, keeps only items from the last N days, applies the compliance keyword filter, normalises timestamps to Beijing time, and writes raw entries to data.js.
3. scripts/digest.py – Calls the configured LLM (via scripts/llm.py) to:
   • Summarise each sector’s recent items into 3‑5 Chinese bullet points.
   • Translate the original English titles.
   • Attach the source URL to each bullet. The enriched data overwrites data.js.
4. index.html – A single‑file static dashboard that reads data.js in the browser and renders the sector panels, bilingual lists, and refresh button.
5. server.py – A tiny http.server‑based web server that serves index.html and exposes /api/refresh which runs the fetch + digest steps in a subprocess.

---

Configuration details

File	Purpose
llm.config.json	Choose the LLM provider. { "provider": "claude-cli" } uses a locally‑installed Claude subscription (requires claude login). Or use { "provider": "api", "api": { "base_url": "…", "api_key": "…", "model": "…" } } for any OpenAI‑compatible endpoint.
sources.json	Editable list of news sources. Adding a source is as simple as appending a JSON object with name, hint (sector tag), type (rss), and url.
Environment variables (optional)	HTTPS_PROXY/HTTP_PROXY for proxying, FETCH_TIMEOUT to increase per‑source timeout, PYTHONIOENCODING if needed.

---

Who it’s for

• Individual investors in China’s A‑share market who need a quick, Chinese‑language snapshot of global industry trends.
• Users who prefer privacy‑first tooling – all data stays on the local machine, no cloud storage.
• Anyone with a Claude subscription or an OpenAI‑compatible API key who wants a zero‑dependency news‑summariser.

---

Limitations & disclaimer (as stated)

• The dashboard shows industry signals, not real‑time price or quantitative market data.
• It is not investment advice; decisions based on the output are the user’s responsibility.
• The tool respects source terms of service by low‑frequency RSS fetching; heavy use may require additional permission.

---

License

MIT – free to use, modify, and redistribute.

---

Bottom line: Investment News is a lightweight, privacy‑preserving Python application that leverages a local or API‑based LLM to turn dozens of international tech‑business feeds into a concise, bilingual daily briefing for Chinese stock‑market investors.