kansoku-trade/kansoku

AI Stock Partner on Your Mac – Local Market Information, Multi-timeframe K-line, AI Intraday Comments and Follow-up Analysis

What it solves

Kansoku is a macOS desktop application designed for stock market analysis that combines real-time data with AI-driven insights. It solves the problem of "black box" AI financial advice by ensuring every AI conclusion is backed by a verifiable evidence chain, archived to prevent retroactive changes, and tracked for accuracy (hit rate).

How it works

The app integrates with the Longbridge CLI to pull market and account data directly to the user's machine. All technical indicators (such as moving averages, MACD, and RS) are calculated locally using TypeScript. Users can configure their own AI models (via API keys or providers like LobeHub Cloud) to perform tasks such as real-time market monitoring, answering questions about charts, and refining research notes. Data and API keys are stored locally in an encrypted SQLite database, ensuring privacy and security.

Who it’s for

It is built for stock traders and investors on macOS who want an AI-powered assistant for market analysis, technical charting, and research management without sacrificing data privacy or transparency.

Highlights

  • Verifiable AI Insights: Every AI conclusion includes a full evidence chain and is archived as a frozen record to track historical accuracy.
  • Local-First Architecture: Market data, technical indicators, and API keys are processed and stored locally; no data leaves the user's machine via intermediate servers.
  • AI-Integrated Charting: AI can read user-drawn lines on charts, suggest key price levels by drawing purple dashed lines directly on the graph, and provide scenario-based predictions (Bull/Base/Bear).
  • SEPA Strategy Dashboard: Includes a built-in dashboard for the Minervini trend template with automatic Buy/Watch List/Avoid conclusions.
  • Model Flexibility: Supports multiple AI providers (OpenAI, Anthropic, Google, DeepSeek) with the ability to assign different models to different tasks (e.g., fast commentary vs. deep research).

Related

  • Project
  • Project
  • Project
  • Project