simonlin1212/Vibe-Research
Vibe-Research: Your Personal Trading Research Agent · A股/美股/港股 的个人投研 Agent:每日复盘、资讯雷达、个股数据、板块中心、我的持仓、研究记录、回测。Vibe-Research 把数据和功能配齐,由你自己的 Agent 驱动投资研究。基于开源的 Codex Harness 打造。
What it solves
Vibe Research is a local financial research workstation designed to transform raw financial data into structured, verifiable research reports. It solves the problem of "hallucinations" and unverifiable claims in AI-driven financial analysis by enforcing a strict evidence-based workflow where every key data point must be backed by a source, a calculation, or a verified document.
How it works
The system operates as a local workbench that connects to various AI runtimes (such as OpenAI Codex, Claude Code, or WorkBuddy/CodeBuddy) or direct model APIs. It uses a three-layer constraint system to ensure accuracy:
- Prompt Layer: Defines financial research disciplines and Standard Operating Procedures (SOPs).
- Execution Layer: Uses sandboxed environments and controlled MCP (Model Context Protocol) tools to limit network and file access.
- Orchestration Layer: Forces a multi-stage research process, requiring evidence citations, deterministic calculations, and conflict detection across sources.
Who it’s for
It is built for financial analysts and researchers who need a professional-grade tool for A-share (Chinese) market research, portfolio tracking, and evidence-based AI assistance without sacrificing data privacy or accuracy.
Highlights
- Six-Stage Research: A structured workflow for A-shares covering company profiles, financials, expectations, valuation, risk, and final reporting.
- Verifiable Outputs: Generates
evidence.jsonandcalculations.jsonalongside reports, ensuring every number is traceable to its origin. - Multi-Runtime Support: Integrates with existing subscriptions like Claude Code and WorkBuddy, or custom API keys.
- Extensive Data Integration: Accesses 117 data endpoints across CN, US, and HK markets for quotes, financials, and industry signals.
- Local-First Privacy: Keeps original research documents on the local machine, sending only relevant snippets to the AI.
- Deterministic Backtesting: Provides a dedicated engine for quantitative verification of research hypotheses.
Related
- Project
- Project
- Project
- Project
- Project