edinetdb/dexter-jp

What it solves

Dexter JP automates the complex process of researching Japanese stocks. Instead of manually gathering data from multiple sources and performing calculations, users can ask high-level analytical questions and receive a structured report based on real-time financial data, regulatory filings, and market trends.

How it works

It operates as an autonomous agent using a LangChain-based loop that plans, selects tools, executes tasks, and verifies results. It integrates several key data sources:

• Financial Data: Uses EDINET DB for financial statements, key ratios, and company info.
• Regulatory Filings: Reads annual reports (有報) and large shareholding reports.
• Market Data: Integrates J-Quants for official Tokyo Stock Exchange price data.
• Web Search: Optional integration with Exa, Perplexity, or Tavily for current events.

The system features a "meta-tool" for financials that uses its own internal LLM to route natural language queries to specific sub-tools. It also includes a skill system for complex workflows like DCF valuation and a customizable rules system (RULES.md) to define specific investment styles or analysis guidelines.

Who it’s for

It is designed for investors, financial analysts, and researchers focusing on the Japanese equity market who want to automate the data collection and preliminary analysis phase of their research.

Highlights

• Autonomous Planning: Can handle complex prompts (e.g., comparing two companies across multiple metrics) by creating its own research plan.
• Multi-Channel Access: Can be deployed as a CLI tool or integrated into Slack, Discord, and LINE.
• Context Compression: Automatically summarizes large datasets during long sessions to save LLM tokens while maintaining analytical consistency.
• Flexible LLM Support: Compatible with OpenAI, Anthropic, Google, xAI, OpenRouter, and local models via Ollama.
• Specialized Japanese Market Logic: Includes DCF valuation tailored to Japanese government bond yields and TSE-specific contexts.