Original Summary

I've been building this alone for about 6–8 months. It's not public and there's no link. I just wanted to show it somewhere and hear what people think before I decide what to do with it. The problem I started with: Every trading tool either costs money, has limitations, or doesn’t offer what you want. That’s why I built my own, which combines all the features in one place, with its own chart data including full SIP volume, and a custom-developed search engine that lets you access any website without relying on external search services for the AI. The same applies to the chart tools and chart data. It also includes a seamless integration with AI, allowing you to monitor, review, and analyze everything, so you only pay for what you actually use. Simply put: Here, you can get the latest news, store your own data, analyze it with AI, and test strategies and patterns. This way, you can completely organize your trading life and no longer have to use 20 external tools. The chat, the cockpit At the bottom you pick which AI model you're talking to. Adding a new one takes a few env variables and no code change. DeepSeek, GPT, Grok and Gemini variants are wired up right now. Every area of the app has its own role file with hard instructions and tool definitions, so the models don't turn into yes-men and don't invent numbers. The chat has a few modes: Live search: the biggest module in the project, and probably the best part. It can reach basically any site on the web in real time, and the whole stack is self-hosted, no search API, no external service: A self-hosted search engine queries five result pages in parallel, then three different scrapers take turns depending on how hostile the site is, and a local reranker model sorts what's actually relevant. The last tier runs in the background so it never blocks your answer. A filter layer drops junk domains before the model ever sees them, and every source is listed under the answer so you can check where it came from. It also renders a mini chart for whatever ticker you're discussing. Personal / data mode: here the AI gets access to your stuff: your journal and your trades. It can read, add entries when you ask, and delete. More below. Earnings mode: past earnings for a company plus a short AI read on the reaction. The journal, and the AI that refuses to answer The journal itself is deliberately simple: notes, mistakes, results, whatever you want to write down, in a tree. The AI can read it, edit it and add to it on request. The Journal AI is the interesting half. It sees every trade, mistake and result you've logged, and its only job is statistics, finding the patterns you repeat, especially the bad ones. It has hard thresholds built in: Under 30 completed trades per group it refuses to give you a statistic, and says so. It'll still answer factual and search questions. Counts stay flagged provisional until 150 completed trades. That's deliberate. With twelve trades

  • 情报分类：工作与职业机会
  • 分类依据：内容涉及招聘、求职或职业发展
  • 信息来源：Reddit · SideProject
  • 发布时间：2026/9/15 23:36:11