monarchjuno/tradingcodex
Turn Codex into your investment workflow team
What it solves
TradingCodex transforms fragmented investment tools into a structured research workflow. It prevents the loss of research history by creating a local-first workspace where evidence, reports, and decision-memory artifacts are saved and inspectable after a chat session ends.
How it works
It operates as an "investment operating system" for Codex, using a team of bounded, fixed-role agents to gather and challenge evidence based on natural language requests. The system is organized into four layers:
- Native Codex task: Handles the initial request and coordinates specialist profiles.
- Workspace: A local folder containing research, reports, and source snapshots.
- Django service: Manages artifact identity, policy, and execution rules.
- Read-only viewer: A web interface for browsing the library and system posture.
Who it’s for
Investment researchers and traders who want a rigorous, evidence-backed research process with a permanent, versionable record of their analysis and strict guardrails between research and trade execution.
Highlights
- Local-first research: Saves all handoffs, evidence, and reports to a user-owned workspace.
- Evidence-backed workflows: Uses a Head Manager to select specialist roles to synthesize and revise research.
- Decision memory: Allows users to replay past decisions to distinguish decision quality from outcome quality.
- Execution guardrails: Strictly separates research and planning from broker authority and order actions.
- OpenBB integration: Optional support for OpenBB as a data source for evidence gathering.