Skip to content

dewilkinson/cobalt-multi-agent

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

203 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Project Cobalt: CMA (Cobalt Multiagent)

A high-integrity, multi-agent intelligence system for individual SMC traders.

Project Cobalt is an agentic framework designed to bridge the gap between raw market data and high-conviction Smart Money Concepts (SMC) analysis. Built on a modular "Blackboard" architecture, Cobalt uses specialized AI agents to scout liquidity, analyze market structure, and deliver graded trade setups directly into your Obsidian vault.


Acknowledgments: This project is a customized iteration built upon the excellent DeerFlow Quickstart framework by ByteDance. We are deeply grateful to its original authors and the open-source community that made the underlying architecture possible.

Contact: D. Wilkinson (dwilkins@bluesec.ai)


Core Architecture: The CMA Framework

Cobalt operates through the CMA (Cobalt Multiagent) logic, which decentralizes the analytical process to ensure data integrity and prevent AI hallucination.

  • VLI (VibeLink Interface): The primary command layer. Interprets natural language "vibes" and context to coordinate agent workflows.
  • The Scout: A hardened data retrieval agent. Connects to brokerage APIs (SnapTrade) to fetch high-fidelity OHLC and trade history.
  • The Analyst: The logic engine. Scans for Liquidity Sweeps, Market Structure Shifts (MSS), and Fair Value Gaps (FVG).
  • The Hub (Obsidian): The local "Source of Truth" where all agent dispatches are stored, archived, and audited.

Security & Philosophy

Cobalt is built with a Security-First mindset:

  • Read-Only by Design: This system has zero execution capability. It cannot place trades or move funds. You are the only "Human-in-the-Loop."
  • Asynchronous Git Bridge: Data moves from the cloud to your local machine via a private Git buffer, allowing for full auditability and offline persistence.
  • Zero-Vault Secrets: No API keys are ever stored in your Markdown files. All credentials reside in encrypted environment variables.

️ License

Project Cobalt is released under the Apache License 2.0.

  • Individual Investors: Free to use, modify, and study for personal trading activities.
  • Commercial Entities: Institutional use, for-profit redistribution, or commercial integration is strictly prohibited.

Getting Started

  1. Clone the Repo: Initialize your Obsidian vault as a Git repository.
  2. Configure Railway: Deploy the Scout agent to Railway and input your SnapTrade credentials as environment variables.
  3. Define your SOP: Set your SMC grading thresholds in the VLI configuration.
  4. Open Obsidian: Use the Obsidian Git plugin to begin receiving Dispatches.

Toolbox: Scripts & Utilities

The scripts/ directory contains various tools to manage and debug the system:

  • scripts/vli/: Essential VLI lifecycle tools (e.g., vli_refresh.sh for emergency system restarts).
  • scripts/diagnostic/: High-fidelity scanners and verification tools for the data pipeline.
  • scripts/utils/: Maintenance utilities like cache cleanup and UI patching.
  • scripts/archive/: Historical logs, test results, and deprecated script iterations.
  • scripts/dev/: Simulation and development-phase logic for testing new agent strategies.

About

Project Cobalt: A multiagent,smart-money-concepts based analysis platform for retail traders. Cobalt coordinates and distributes workloads across a network of LLMs and fixed function python modules to ensure accurate synthesis based on live news and market data. Handles broker connections, full quant analysis, trade management and journaling.

Topics

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • HTML 54.4%
  • Python 27.8%
  • TypeScript 12.5%
  • JavaScript 2.5%
  • Shell 1.7%
  • CSS 1.1%