Building a "Remembering" AI Trading Agent with Python, LangGraph, and Obsidian

#ai #machinelearning #python #opensource

Hello DEV community!

I'm excited to share an open-source project I've been working on: TradingAgents-Dashboard.
It's a Dockerized AI trading assistant that not only analyzes the market but remembers your insights forever using a local knowledge base (RAG).

The Problem: "Stateless" AI

Most AI trading bots today are "stateless". They run an analysis, give you a result, and then forget everything the moment you close the terminal.

"Wait, didn't we decide last week that inflation correlates with this stock?"
"Where is that news link I saw yesterday?"

As a developer and trader, I wanted an agent that grows smarter over time, just like a human analyst.

The Solution: AI + Obsidian (RAG)

I built a dashboard wrapping the TradingAgents framework, adding a persistent memory layer using Obsidian.

Github Repo: jiwoomap/TradingAgents-Dashboard

How it works:

Analyze: Agents (Bull, Bear, Risk Manager) debate market conditions using LangGraph.
Persist: All insights and debates are auto-saved to your local Obsidian Vault as Markdown files.
Recall (RAG): Before making a new decision, the agents search your vault (via ChromaDB) to retrieve past lessons and context.

Tech Stack

Framework: LangChain / LangGraph (Multi-Agent Orchestration)
UI: Streamlit (Web Dashboard)
Database: ChromaDB (Vector Store for RAG)
Memory: Obsidian (Markdown-based Knowledge Base)
Infrastructure: Docker & Docker Compose

Key Features

Interactive Debate UI: Watch the "Bull" and "Bear" agents fight it out in real-time.
Fact Checker: Prevents hallucinations by validating news URLs (200 OK checks).
Dockerized: Get started in 1 minute with docker-compose up.
Data Sovereignty: Your financial data and strategies live on your disk, not in a cloud database.