monarchjuno/tradingcodex

What it solves

TradingCodex provides a structured, research-first investment analysis workflow. It prevents the common issue of generic AI chat responses by using a specialized team of agents that collaborate to challenge evidence and build a persistent, reusable knowledge base for financial decision-making.

How it works

The system uses a "Head Manager" agent that dynamically selects the smallest necessary group of specialist agents (such as Fundamental, Technical, Macro, and Risk Managers) based on the the user's specific question. Instead of a fixed sequence, the workflow is a loop where agents produce evidence and artifacts that are stored in a local, inspectable workspace. This workspace includes a Research Memory for evidence, a Knowledge Wiki for background, an Investment Brain for inquiry frameworks, and Decision Memory for postmortems and lessons.

Who it’s for

It is designed for investors and researchers who want a local-first, evidence-based approach to investment analysis without granting the AI direct execution authority over their brokerage accounts.

Highlights

• Dynamic Agent Selection: The team composition changes based on the mandate, avoiding redundant processing.
• Persistent Workspace: Research, wikis, and decision history are stored as versionable files rather than hidden in chat logs.
• Research-First Guardrails: A strict separation between research and broker action ensures that AI-generated insights never trigger trades automatically.
• Specialized Roles: Includes nine distinct investment specialist agents, including a Judgment Reviewer to independently challenge evidence.