daily_stock_analysis: An Open-Source LLM Stock Analyst That Pushes Daily Reports via GitHub Actions
> Source: ZhuLinsen/daily_stock_analysis · Python · Stars today: 505 · License: MIT > In one sentence: Fork the repo, fill in a few secrets, and every trading day at 6 PM your WeChat Work / Feishu / Telegram receives an LLM-generated decision report for your watchlist.
The Scenario
Imagine a watchlist spanning six markets: Kweichow Moutai (A-shares), Tencent (Hong Kong), Apple (US), Toyota (Japan), Samsung (Korea), and TSMC (Taiwan). After each market close, you want to know: what happened today? Should you act? Any risks?
The traditional approach means opening one app for A-shares, another for HK/US stocks, Yahoo for Japan/Korea/Taiwan, plus Weibo, Twitter, and Bloomberg for news. By the time you finish, the day is over.
daily_stock_analysis offers another path: fork the repository, add your stock codes and API keys, and set up a GitHub Actions scheduled task. Every trading day at 18:00, your WeChat Work / Feishu / Telegram / Discord / Slack / email receives a decision dashboard containing core conclusions, scores, trends, buy/sell levels, risk alerts, catalysts, and an action checklist.
Zero server cost, zero maintenance, and setup takes about five minutes after forking.
Full-Pipeline Breakdown
The project's most notable aspect isn't any single feature — it's that it connects the entire chain of "LLM-based stock analysis."
Data Layer: Multi-Source Market Data
| Market | Sources | |--------|---------| | A-shares | AkShare, Tushare, Pytdx, Baostock | | HK / US | YFinance, Longbridge | | Japan / Korea / Taiwan | YFinance | | ETFs | Full market support |
Free sources (AkShare, Baostock, YFinance) work with zero configuration, but the docs honestly warn they suffer from upstream rate limiting. For long-term scheduled runs, token-based sources like Tushare or Longbridge are recommended.
News Layer: Multi-Source Search Aggregation
| Type | Services | |------|----------| | AI search | Anspire, Tavily, Bocha, Brave, MiniMax | | Search engines | SerpAPI | | Self-hosted | SearXNG | | Social sentiment | Stock Sentiment API (Reddit / X / Polymarket, US stocks only) |
News sources significantly affect the quality of sentiment, announcements, events, and catalysts. The docs explicitly recommend configuring at least one search service — far more honest than typical "works out of the box" marketing.
Strategy Layer: 15 Built-In Strategies
| Type | Coverage | |------|----------| | Technical | Moving averages, Chan theory, Elliott waves, trends | | Event-driven | Hot topics, events | | Fundamental | Growth, expectations | | Agent strategies | Multi-turn follow-up Q&A via Web/Bot/API |
The agent-based Q&A is a highlight: instead of single-shot answers, you can ask "How is Moutai trending lately?" and then follow up with "What about its valuation?" — the agent continues within context.
LLM Layer: Model-Agnostic
Supports Gemini, OpenAI-compatible APIs (including DeepSeek, Qwen), Claude, local models via Ollama, and aggregators like Modelflare and AIHubMix. You can run daily analysis on cheap models and reserve the strongest model for critical decisions.
Delivery Layer: 7 Channels
WeChat Work, Feishu, Telegram, Discord, Slack, email, plus a web workbench and desktop client. What gets pushed isn't just text — it's a decision dashboard with scores, trend charts, buy/sell levels, and risk alerts.
Scheduling Layer: Free GitHub Actions
This is the key to "zero cost": GitHub Actions' free tier is sufficient for personal repos. Runs automatically at 18:00 Beijing time on trading days, skipping non-trading days (including A-share/HK/US holidays). Fork, configure secrets, enable Actions — done.
Academic Backing
The README links to an arXiv paper: 2608.26990. Having academic support suggests the author did serious methodological research rather than just writing code on a whim.
Honest Engineering Docs
The documentation quality is first-tier among trending GitHub projects:
1. Free sources explicitly flagged as unstable — no pretense 2. Dedicated docs for data source priority and fallback rules 3. A dedicated page on market support boundaries and potentially missing fields 4. Clear trading-day check logic (skips non-trading days by default, force-run possible) 5. Checkpoint/resume support for interrupted large-batch analyses
This "documenting the pitfalls openly" style is far more credible than "one-click deploy, works out of the box" projects.
Why It Matters
1. A Full-Pipeline LLM Application Reference
Most LLM projects are either API wrappers or pure data processing. This project connects the entire chain — data collection → news aggregation → LLM analysis → decision report → multi-channel push → scheduled dispatch — with multiple interchangeable sources/services at each step. Even outside finance, the architecture (multi-source aggregation, model-agnostic design, multi-channel push, GitHub Actions scheduling) transfers directly to other domains.
2. The Zero-Cost Automation Paradigm
The GitHub Actions + fork pattern is common in open source (RSS bots, GitHub Pages), but applying it to "scheduled LLM analysis + push" is clever:
- No server needed
- No Docker
- No ops
- Fork-and-run; configure via secrets
- Individual investors with a watchlist who want daily AI reports — fork and run
- LLM application developers studying full-pipeline design
- Quant enthusiasts — 15 strategies + multi-market data as a validation starting point
- GitHub Actions learners — a textbook fork-and-run implementation
- GitHub: https://github.com/ZhuLinsen/daily_stock_analysis
- Full guide: docs/full-guide.md
- arXiv paper: https://arxiv.org/abs/2608.26990
- Bilibili tutorial: https://www.bilibili.com/video/BV11FEb66EXG/
This fork-and-run model may be an important path for bringing LLM applications to the masses.
3. Real Engineering Complexity
Adapting data sources across 6 markets, unifying 15 strategies behind one interface, abstracting 7 push channels — this is not a weekend project. If you build any "multi-source data + LLM analysis" system, the architecture is worth reading at the source level.
Who Should Use It
One Caveat
A small note atop the README: the project has sponsors (Modelflare, SerpApi), and sponsor API keys are marked "recommended." This isn't hidden commercial placement — the sponsorship is disclosed and all keys are optional; you can run entirely on free sources. Still, "recommended" doesn't mean "optimal," so compare options.
---
Links