Is this free GitHub AI stock analyst actually useful?

追加: 2026-05-30

[00:00:00]Almost 40,000 developers starred this repo. Almost the exact same number forked it. And that one detail tells you everything about how it actually works.

[00:00:10]Here's the thing. This isn't an app you download or a dashboard you log into.

[00:00:15]It's a repository you fork. You drop one AI model key into your fork secrets, switch on GitHub actions, and from then on their own servers do the work. Every weekday right around market close, it wakes up, pulls the latest prices and news for whatever's on your watch list, runs all of it through a large language model, then pushes a clean buy or sell call straight to your Telegram. There's no server to rent and no laptop left running overnight. It's MIT licensed and the whole thing is built from the ground up to live on somebody else's free compute.

[00:00:50]Enjoying this so far? Subscribe to the YouTube channel and grab the free daily newsletter at indiehacker.news.

[00:00:57]New deep dive every weekday.

[00:01:00]Let me actually walk you through the repo because the page explains the hype better than I can. Up top, you get the badges and a short pitch in both Chinese and English since this started life as a Chinese project and grew into one of the most forked finance repos on the whole site. Right below that is a live preview of the web workspace and then the feature matrix that lays out exactly what each part of the system does.

[00:01:25]Keep going and you reach the part that actually matters. The 5-minute setup where you fork the project and then flip the schedule job on once your keys are in. And near the bottom, it shows you a real sample of what lands on your phone, a dated dashboard with a verdict and a score for every stock you follow.

[00:01:44]So, how is this different from every other stock tool out there? Most of them fall into two camps. Paid screeners and robo advisors charge you a monthly fee just to log in and look around. Trading bots go the other way. You host them yourself and they place real trades with real money on your behalf. This one sits in a third spot that barely existed before. It never touches your brokerage, so it physically cannot lose your money by pulling the trigger and there's nothing for you to keep online either.

[00:02:16]You also don't have to remember to open anything. The report just arrives the moment the market closes.

[00:02:22]And the raw numbers here are honestly a little absurd. That ratio is the part to pay attention to because people don't just bookmark this project, they clone it and run it themselves.

[00:02:34]Okay, so what actually shows up on your phone? The centerpiece is the decision report. For every stock on your list, you get a one-word verdict and a score from 0 to 100 and underneath that sit the suggested entry and exit levels plus a plain action checklist. To build all that, it chews through a genuinely deep pile of raw data. Everything from real-time quotes and candlestick history down to capital flow and even chip distribution, then layers fresh news and company filings on top. And it does all of this for A shares and Hong Kong names right alongside American tickers and ETFs in the very same run. Next to the per stock view, there's also a daily market review that zooms out to the big indices and shows you how breadth and the leading sectors shook out.

[00:03:24]Want to be featured? Pitch your open-source repo at indiehacker.news/apply and get in front of thousands of builders.

[00:03:33]There's a whole second half to this that has nothing to do with the scheduled job. You can open a chat and just ask it things and sitting behind that chat are 15 built-in strategies. Anything from a plain moving average crossover up to Elliott Wave and event-driven setups and you can keep drilling in with follow-up questions. There's a full workbench, too, a web and desktop app where you run your own analysis by hand and dig back through your history with backtesting and portfolio tracking sitting right there in light or dark mode. A recent version added a proper alerts engine that watches indicators like RSI and MACD and pings you when one of your holdings crosses the line you set. And you can hand it your watch list by pasting in a screenshot or spreadsheet.

[00:04:20]It reads the tickers right out of the image.

[00:04:23]Architecturally, it's a Python core with a TypeScript desktop and web layer on top. The clever part is the model routing. It leans on light LLM, so you can point it at a hosted name like Gemini or Claude one day and a local Ollama model the next or really any OpenAI compatible endpoint and swap between them without touching code. For market data, it doesn't lean on a single provider, either. It pulls from a whole rotation of sources like to share and Yahoo Finance with automatic failover when one of them goes down. News and sentiment come in through their own set of search APIs, and the scheduler itself is just a cron job firing Monday through Friday with a deliberate random delay baked in so that tens of thousands of forks don't all hammer the same data sources at the very same second.

[00:05:17]Now, if you'd rather not deal with a scheduled job in the cloud at all, running this on your own machine is the short sequence you can see on screen.

[00:05:25]You pull it down and point it at your keys and you're going. From there, a few flags let you narrow it to a single ticker or boot up the full web interface instead. But honestly, the fork and forget route is the whole appeal here.

[00:05:39]The local path is really there for folks who want it humming on a private box or inside Docker.

[00:05:45]The latest release, version 3.18 is mostly about reliability. The alert center got a serious upgrade. It now runs its evaluations quietly in the background and remembers cool downs so a noisy ticker can't spam you all day. You can also pick a strategy per analysis now like event driven or growth quality instead of getting one generic take every time. On the data side, they wired in two more American sources Fin Hub and Alpha Vantage with a longer fallback chain so US tickers break less often and the model layer got tougher with automatic retries inside a single request and pricing fallbacks for brand new models it hasn't seen before.

[00:06:28]So here's the honest read. This is unapologetically built for Chinese markets first. The docs and the defaults all lead with A shares and while it does handle American tickers, that support is younger and clearly second priority. The word free deserves an asterisk too because the compute is free but the intelligence isn't. You bring your own model key and you pay for the tokens it burns unless you stay inside a free tier and exactly like the bold disclaimer warns this is a research toy and not a financial advisor since a confident sounding model can still be confidently wrong about your money. But if you're the kind of builder who wants a market briefing waiting every evening without running a single machine of your own, then pointing a forked repo at a cheap model key is a genuinely clever trick.

[00:07:17]The same author already ships two sister projects right beside it. One for hunting candidate stocks and one for evolving strategies. So this might only be the first piece of the puzzle they hand you.