A trading review knowledge base powered by LLMs, this project provides a CLI toolchain for traders and researchers to organize raw data, wikis, graphs, and facts into a searchable, verifiable, and iterative trading research system. It offers multi-source retrieval, automated workflows for prediction and validation, and advanced features for knowledge governance, aiming to enhance trading analysis and decision-making.

How It Works

The system leverages a multi-source Retrieval Augmented Generation (RAG) approach, integrating various data types including formal wikis, raw text, knowledge graphs, structured facts, long-term memory ("brain"), and market SQL data. A core component is the Codex CLI toolchain, which orchestrates data ingestion, retrieval, and analysis. A novel Temporal Facts v1 layer (data/facts/temporal_edges.jsonl) tracks facts with time-sensitive or verifiable states (e.g., supersedes, contradictedBy), allowing for auditing and preventing outdated information from polluting current answers. This design enables a live, automated knowledge base for continuous research and validation.

Quick Start & Requirements

• Install: Clone the repository and run npm install.
• Prerequisites: Node.js is required for the npm commands. An optional PostgreSQL database can be configured for market data (stock_daily_sql) integration.
• Usage: Execute commands via npm run codex:ingest -- <command>, such as search, smart-search, ask, daily-loop, company-research, or temporal-facts audit.
• Documentation: Detailed guides are available in docs/temporal-facts-v1.md and docs/多源检索RAG完整流程.md.

Highlighted Details

• Multi-Source RAG: Seamlessly queries across wiki, raw, graph, facts, brain, and stock_daily_sql sources.
• Temporal Facts v1: Manages and audits time-sensitive, verifiable, or superseded facts, enhancing data integrity.
• Automated Workflows: Includes daily-loop for pre-market prediction and post-market validation, and company-research for in-depth company analysis.
• Brain Memory: Facilitates long-term error correction and self-training through remember, status, and resolve commands.

Maintenance & Community

The project lists @sowelswl as a collaborator. 上海玄玑灵析科技 provided support for personal growth and technical assistance. The author's WeChat public account (ymj0418) is also mentioned.

Licensing & Compatibility

The project is licensed under the GNU General Public License v3.0 (GPL-3.0). This is a strong copyleft license, meaning derivative works must also be licensed under GPL-3.0. Commercial use or integration with closed-source systems may be subject to significant restrictions.

Limitations & Caveats

The main branch is optimized for CLI automation; users preferring a desktop application should utilize historical release artifacts. The system relies on a local, securely configured knowledge base workspace, and market data integration requires local PostgreSQL credentials that are not stored in the repository.